xref: /openbmc/linux/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c (revision 023e41632e065d49bcbe31b3c4b336217f96a271)
1 /*
2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
3  *
4  * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
44 #include <linux/if.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <net/addrconf.h>
64 #include <net/bonding.h>
65 #include <linux/uaccess.h>
66 #include <linux/crash_dump.h>
67 #include <net/udp_tunnel.h>
68 
69 #include "cxgb4.h"
70 #include "cxgb4_filter.h"
71 #include "t4_regs.h"
72 #include "t4_values.h"
73 #include "t4_msg.h"
74 #include "t4fw_api.h"
75 #include "t4fw_version.h"
76 #include "cxgb4_dcb.h"
77 #include "srq.h"
78 #include "cxgb4_debugfs.h"
79 #include "clip_tbl.h"
80 #include "l2t.h"
81 #include "smt.h"
82 #include "sched.h"
83 #include "cxgb4_tc_u32.h"
84 #include "cxgb4_tc_flower.h"
85 #include "cxgb4_ptp.h"
86 #include "cxgb4_cudbg.h"
87 
88 char cxgb4_driver_name[] = KBUILD_MODNAME;
89 
90 #ifdef DRV_VERSION
91 #undef DRV_VERSION
92 #endif
93 #define DRV_VERSION "2.0.0-ko"
94 const char cxgb4_driver_version[] = DRV_VERSION;
95 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
96 
97 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
98 			 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
99 			 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
100 
101 /* Macros needed to support the PCI Device ID Table ...
102  */
103 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
104 	static const struct pci_device_id cxgb4_pci_tbl[] = {
105 #define CXGB4_UNIFIED_PF 0x4
106 
107 #define CH_PCI_DEVICE_ID_FUNCTION CXGB4_UNIFIED_PF
108 
109 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
110  * called for both.
111  */
112 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
113 
114 #define CH_PCI_ID_TABLE_ENTRY(devid) \
115 		{PCI_VDEVICE(CHELSIO, (devid)), CXGB4_UNIFIED_PF}
116 
117 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
118 		{ 0, } \
119 	}
120 
121 #include "t4_pci_id_tbl.h"
122 
123 #define FW4_FNAME "cxgb4/t4fw.bin"
124 #define FW5_FNAME "cxgb4/t5fw.bin"
125 #define FW6_FNAME "cxgb4/t6fw.bin"
126 #define FW4_CFNAME "cxgb4/t4-config.txt"
127 #define FW5_CFNAME "cxgb4/t5-config.txt"
128 #define FW6_CFNAME "cxgb4/t6-config.txt"
129 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
130 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
131 #define PHY_AQ1202_DEVICEID 0x4409
132 #define PHY_BCM84834_DEVICEID 0x4486
133 
134 MODULE_DESCRIPTION(DRV_DESC);
135 MODULE_AUTHOR("Chelsio Communications");
136 MODULE_LICENSE("Dual BSD/GPL");
137 MODULE_VERSION(DRV_VERSION);
138 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
139 MODULE_FIRMWARE(FW4_FNAME);
140 MODULE_FIRMWARE(FW5_FNAME);
141 MODULE_FIRMWARE(FW6_FNAME);
142 
143 /*
144  * The driver uses the best interrupt scheme available on a platform in the
145  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
146  * of these schemes the driver may consider as follows:
147  *
148  * msi = 2: choose from among all three options
149  * msi = 1: only consider MSI and INTx interrupts
150  * msi = 0: force INTx interrupts
151  */
152 static int msi = 2;
153 
154 module_param(msi, int, 0644);
155 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
156 
157 /*
158  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
159  * offset by 2 bytes in order to have the IP headers line up on 4-byte
160  * boundaries.  This is a requirement for many architectures which will throw
161  * a machine check fault if an attempt is made to access one of the 4-byte IP
162  * header fields on a non-4-byte boundary.  And it's a major performance issue
163  * even on some architectures which allow it like some implementations of the
164  * x86 ISA.  However, some architectures don't mind this and for some very
165  * edge-case performance sensitive applications (like forwarding large volumes
166  * of small packets), setting this DMA offset to 0 will decrease the number of
167  * PCI-E Bus transfers enough to measurably affect performance.
168  */
169 static int rx_dma_offset = 2;
170 
171 /* TX Queue select used to determine what algorithm to use for selecting TX
172  * queue. Select between the kernel provided function (select_queue=0) or user
173  * cxgb_select_queue function (select_queue=1)
174  *
175  * Default: select_queue=0
176  */
177 static int select_queue;
178 module_param(select_queue, int, 0644);
179 MODULE_PARM_DESC(select_queue,
180 		 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
181 
182 static struct dentry *cxgb4_debugfs_root;
183 
184 LIST_HEAD(adapter_list);
185 DEFINE_MUTEX(uld_mutex);
186 
187 static void link_report(struct net_device *dev)
188 {
189 	if (!netif_carrier_ok(dev))
190 		netdev_info(dev, "link down\n");
191 	else {
192 		static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
193 
194 		const char *s;
195 		const struct port_info *p = netdev_priv(dev);
196 
197 		switch (p->link_cfg.speed) {
198 		case 100:
199 			s = "100Mbps";
200 			break;
201 		case 1000:
202 			s = "1Gbps";
203 			break;
204 		case 10000:
205 			s = "10Gbps";
206 			break;
207 		case 25000:
208 			s = "25Gbps";
209 			break;
210 		case 40000:
211 			s = "40Gbps";
212 			break;
213 		case 50000:
214 			s = "50Gbps";
215 			break;
216 		case 100000:
217 			s = "100Gbps";
218 			break;
219 		default:
220 			pr_info("%s: unsupported speed: %d\n",
221 				dev->name, p->link_cfg.speed);
222 			return;
223 		}
224 
225 		netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
226 			    fc[p->link_cfg.fc]);
227 	}
228 }
229 
230 #ifdef CONFIG_CHELSIO_T4_DCB
231 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
232 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
233 {
234 	struct port_info *pi = netdev_priv(dev);
235 	struct adapter *adap = pi->adapter;
236 	struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
237 	int i;
238 
239 	/* We use a simple mapping of Port TX Queue Index to DCB
240 	 * Priority when we're enabling DCB.
241 	 */
242 	for (i = 0; i < pi->nqsets; i++, txq++) {
243 		u32 name, value;
244 		int err;
245 
246 		name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
247 			FW_PARAMS_PARAM_X_V(
248 				FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
249 			FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
250 		value = enable ? i : 0xffffffff;
251 
252 		/* Since we can be called while atomic (from "interrupt
253 		 * level") we need to issue the Set Parameters Commannd
254 		 * without sleeping (timeout < 0).
255 		 */
256 		err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
257 					    &name, &value,
258 					    -FW_CMD_MAX_TIMEOUT);
259 
260 		if (err)
261 			dev_err(adap->pdev_dev,
262 				"Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
263 				enable ? "set" : "unset", pi->port_id, i, -err);
264 		else
265 			txq->dcb_prio = enable ? value : 0;
266 	}
267 }
268 
269 int cxgb4_dcb_enabled(const struct net_device *dev)
270 {
271 	struct port_info *pi = netdev_priv(dev);
272 
273 	if (!pi->dcb.enabled)
274 		return 0;
275 
276 	return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
277 		(pi->dcb.state == CXGB4_DCB_STATE_HOST));
278 }
279 #endif /* CONFIG_CHELSIO_T4_DCB */
280 
281 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
282 {
283 	struct net_device *dev = adapter->port[port_id];
284 
285 	/* Skip changes from disabled ports. */
286 	if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
287 		if (link_stat)
288 			netif_carrier_on(dev);
289 		else {
290 #ifdef CONFIG_CHELSIO_T4_DCB
291 			if (cxgb4_dcb_enabled(dev)) {
292 				cxgb4_dcb_reset(dev);
293 				dcb_tx_queue_prio_enable(dev, false);
294 			}
295 #endif /* CONFIG_CHELSIO_T4_DCB */
296 			netif_carrier_off(dev);
297 		}
298 
299 		link_report(dev);
300 	}
301 }
302 
303 void t4_os_portmod_changed(struct adapter *adap, int port_id)
304 {
305 	static const char *mod_str[] = {
306 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
307 	};
308 
309 	struct net_device *dev = adap->port[port_id];
310 	struct port_info *pi = netdev_priv(dev);
311 
312 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
313 		netdev_info(dev, "port module unplugged\n");
314 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
315 		netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
316 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
317 		netdev_info(dev, "%s: unsupported port module inserted\n",
318 			    dev->name);
319 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
320 		netdev_info(dev, "%s: unknown port module inserted\n",
321 			    dev->name);
322 	else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
323 		netdev_info(dev, "%s: transceiver module error\n", dev->name);
324 	else
325 		netdev_info(dev, "%s: unknown module type %d inserted\n",
326 			    dev->name, pi->mod_type);
327 
328 	/* If the interface is running, then we'll need any "sticky" Link
329 	 * Parameters redone with a new Transceiver Module.
330 	 */
331 	pi->link_cfg.redo_l1cfg = netif_running(dev);
332 }
333 
334 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
335 module_param(dbfifo_int_thresh, int, 0644);
336 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
337 
338 /*
339  * usecs to sleep while draining the dbfifo
340  */
341 static int dbfifo_drain_delay = 1000;
342 module_param(dbfifo_drain_delay, int, 0644);
343 MODULE_PARM_DESC(dbfifo_drain_delay,
344 		 "usecs to sleep while draining the dbfifo");
345 
346 static inline int cxgb4_set_addr_hash(struct port_info *pi)
347 {
348 	struct adapter *adap = pi->adapter;
349 	u64 vec = 0;
350 	bool ucast = false;
351 	struct hash_mac_addr *entry;
352 
353 	/* Calculate the hash vector for the updated list and program it */
354 	list_for_each_entry(entry, &adap->mac_hlist, list) {
355 		ucast |= is_unicast_ether_addr(entry->addr);
356 		vec |= (1ULL << hash_mac_addr(entry->addr));
357 	}
358 	return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
359 				vec, false);
360 }
361 
362 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
363 {
364 	struct port_info *pi = netdev_priv(netdev);
365 	struct adapter *adap = pi->adapter;
366 	int ret;
367 	u64 mhash = 0;
368 	u64 uhash = 0;
369 	bool free = false;
370 	bool ucast = is_unicast_ether_addr(mac_addr);
371 	const u8 *maclist[1] = {mac_addr};
372 	struct hash_mac_addr *new_entry;
373 
374 	ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
375 				NULL, ucast ? &uhash : &mhash, false);
376 	if (ret < 0)
377 		goto out;
378 	/* if hash != 0, then add the addr to hash addr list
379 	 * so on the end we will calculate the hash for the
380 	 * list and program it
381 	 */
382 	if (uhash || mhash) {
383 		new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
384 		if (!new_entry)
385 			return -ENOMEM;
386 		ether_addr_copy(new_entry->addr, mac_addr);
387 		list_add_tail(&new_entry->list, &adap->mac_hlist);
388 		ret = cxgb4_set_addr_hash(pi);
389 	}
390 out:
391 	return ret < 0 ? ret : 0;
392 }
393 
394 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
395 {
396 	struct port_info *pi = netdev_priv(netdev);
397 	struct adapter *adap = pi->adapter;
398 	int ret;
399 	const u8 *maclist[1] = {mac_addr};
400 	struct hash_mac_addr *entry, *tmp;
401 
402 	/* If the MAC address to be removed is in the hash addr
403 	 * list, delete it from the list and update hash vector
404 	 */
405 	list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
406 		if (ether_addr_equal(entry->addr, mac_addr)) {
407 			list_del(&entry->list);
408 			kfree(entry);
409 			return cxgb4_set_addr_hash(pi);
410 		}
411 	}
412 
413 	ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
414 	return ret < 0 ? -EINVAL : 0;
415 }
416 
417 /*
418  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
419  * If @mtu is -1 it is left unchanged.
420  */
421 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
422 {
423 	struct port_info *pi = netdev_priv(dev);
424 	struct adapter *adapter = pi->adapter;
425 
426 	__dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
427 	__dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
428 
429 	return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
430 			     (dev->flags & IFF_PROMISC) ? 1 : 0,
431 			     (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
432 			     sleep_ok);
433 }
434 
435 /**
436  *	cxgb4_change_mac - Update match filter for a MAC address.
437  *	@pi: the port_info
438  *	@viid: the VI id
439  *	@tcam_idx: TCAM index of existing filter for old value of MAC address,
440  *		   or -1
441  *	@addr: the new MAC address value
442  *	@persist: whether a new MAC allocation should be persistent
443  *	@add_smt: if true also add the address to the HW SMT
444  *
445  *	Modifies an MPS filter and sets it to the new MAC address if
446  *	@tcam_idx >= 0, or adds the MAC address to a new filter if
447  *	@tcam_idx < 0. In the latter case the address is added persistently
448  *	if @persist is %true.
449  *	Addresses are programmed to hash region, if tcam runs out of entries.
450  *
451  */
452 static int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
453 			    int *tcam_idx, const u8 *addr, bool persist,
454 			    u8 *smt_idx)
455 {
456 	struct adapter *adapter = pi->adapter;
457 	struct hash_mac_addr *entry, *new_entry;
458 	int ret;
459 
460 	ret = t4_change_mac(adapter, adapter->mbox, viid,
461 			    *tcam_idx, addr, persist, smt_idx);
462 	/* We ran out of TCAM entries. try programming hash region. */
463 	if (ret == -ENOMEM) {
464 		/* If the MAC address to be updated is in the hash addr
465 		 * list, update it from the list
466 		 */
467 		list_for_each_entry(entry, &adapter->mac_hlist, list) {
468 			if (entry->iface_mac) {
469 				ether_addr_copy(entry->addr, addr);
470 				goto set_hash;
471 			}
472 		}
473 		new_entry = kzalloc(sizeof(*new_entry), GFP_KERNEL);
474 		if (!new_entry)
475 			return -ENOMEM;
476 		ether_addr_copy(new_entry->addr, addr);
477 		new_entry->iface_mac = true;
478 		list_add_tail(&new_entry->list, &adapter->mac_hlist);
479 set_hash:
480 		ret = cxgb4_set_addr_hash(pi);
481 	} else if (ret >= 0) {
482 		*tcam_idx = ret;
483 		ret = 0;
484 	}
485 
486 	return ret;
487 }
488 
489 /*
490  *	link_start - enable a port
491  *	@dev: the port to enable
492  *
493  *	Performs the MAC and PHY actions needed to enable a port.
494  */
495 static int link_start(struct net_device *dev)
496 {
497 	int ret;
498 	struct port_info *pi = netdev_priv(dev);
499 	unsigned int mb = pi->adapter->pf;
500 
501 	/*
502 	 * We do not set address filters and promiscuity here, the stack does
503 	 * that step explicitly.
504 	 */
505 	ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
506 			    !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
507 	if (ret == 0)
508 		ret = cxgb4_change_mac(pi, pi->viid, &pi->xact_addr_filt,
509 				       dev->dev_addr, true, &pi->smt_idx);
510 	if (ret == 0)
511 		ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
512 				    &pi->link_cfg);
513 	if (ret == 0) {
514 		local_bh_disable();
515 		ret = t4_enable_pi_params(pi->adapter, mb, pi, true,
516 					  true, CXGB4_DCB_ENABLED);
517 		local_bh_enable();
518 	}
519 
520 	return ret;
521 }
522 
523 #ifdef CONFIG_CHELSIO_T4_DCB
524 /* Handle a Data Center Bridging update message from the firmware. */
525 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
526 {
527 	int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
528 	struct net_device *dev = adap->port[adap->chan_map[port]];
529 	int old_dcb_enabled = cxgb4_dcb_enabled(dev);
530 	int new_dcb_enabled;
531 
532 	cxgb4_dcb_handle_fw_update(adap, pcmd);
533 	new_dcb_enabled = cxgb4_dcb_enabled(dev);
534 
535 	/* If the DCB has become enabled or disabled on the port then we're
536 	 * going to need to set up/tear down DCB Priority parameters for the
537 	 * TX Queues associated with the port.
538 	 */
539 	if (new_dcb_enabled != old_dcb_enabled)
540 		dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
541 }
542 #endif /* CONFIG_CHELSIO_T4_DCB */
543 
544 /* Response queue handler for the FW event queue.
545  */
546 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
547 			  const struct pkt_gl *gl)
548 {
549 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
550 
551 	rsp++;                                          /* skip RSS header */
552 
553 	/* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
554 	 */
555 	if (unlikely(opcode == CPL_FW4_MSG &&
556 	   ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
557 		rsp++;
558 		opcode = ((const struct rss_header *)rsp)->opcode;
559 		rsp++;
560 		if (opcode != CPL_SGE_EGR_UPDATE) {
561 			dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
562 				, opcode);
563 			goto out;
564 		}
565 	}
566 
567 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
568 		const struct cpl_sge_egr_update *p = (void *)rsp;
569 		unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
570 		struct sge_txq *txq;
571 
572 		txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
573 		txq->restarts++;
574 		if (txq->q_type == CXGB4_TXQ_ETH) {
575 			struct sge_eth_txq *eq;
576 
577 			eq = container_of(txq, struct sge_eth_txq, q);
578 			t4_sge_eth_txq_egress_update(q->adap, eq, -1);
579 		} else {
580 			struct sge_uld_txq *oq;
581 
582 			oq = container_of(txq, struct sge_uld_txq, q);
583 			tasklet_schedule(&oq->qresume_tsk);
584 		}
585 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
586 		const struct cpl_fw6_msg *p = (void *)rsp;
587 
588 #ifdef CONFIG_CHELSIO_T4_DCB
589 		const struct fw_port_cmd *pcmd = (const void *)p->data;
590 		unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
591 		unsigned int action =
592 			FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
593 
594 		if (cmd == FW_PORT_CMD &&
595 		    (action == FW_PORT_ACTION_GET_PORT_INFO ||
596 		     action == FW_PORT_ACTION_GET_PORT_INFO32)) {
597 			int port = FW_PORT_CMD_PORTID_G(
598 					be32_to_cpu(pcmd->op_to_portid));
599 			struct net_device *dev;
600 			int dcbxdis, state_input;
601 
602 			dev = q->adap->port[q->adap->chan_map[port]];
603 			dcbxdis = (action == FW_PORT_ACTION_GET_PORT_INFO
604 			  ? !!(pcmd->u.info.dcbxdis_pkd & FW_PORT_CMD_DCBXDIS_F)
605 			  : !!(be32_to_cpu(pcmd->u.info32.lstatus32_to_cbllen32)
606 			       & FW_PORT_CMD_DCBXDIS32_F));
607 			state_input = (dcbxdis
608 				       ? CXGB4_DCB_INPUT_FW_DISABLED
609 				       : CXGB4_DCB_INPUT_FW_ENABLED);
610 
611 			cxgb4_dcb_state_fsm(dev, state_input);
612 		}
613 
614 		if (cmd == FW_PORT_CMD &&
615 		    action == FW_PORT_ACTION_L2_DCB_CFG)
616 			dcb_rpl(q->adap, pcmd);
617 		else
618 #endif
619 			if (p->type == 0)
620 				t4_handle_fw_rpl(q->adap, p->data);
621 	} else if (opcode == CPL_L2T_WRITE_RPL) {
622 		const struct cpl_l2t_write_rpl *p = (void *)rsp;
623 
624 		do_l2t_write_rpl(q->adap, p);
625 	} else if (opcode == CPL_SMT_WRITE_RPL) {
626 		const struct cpl_smt_write_rpl *p = (void *)rsp;
627 
628 		do_smt_write_rpl(q->adap, p);
629 	} else if (opcode == CPL_SET_TCB_RPL) {
630 		const struct cpl_set_tcb_rpl *p = (void *)rsp;
631 
632 		filter_rpl(q->adap, p);
633 	} else if (opcode == CPL_ACT_OPEN_RPL) {
634 		const struct cpl_act_open_rpl *p = (void *)rsp;
635 
636 		hash_filter_rpl(q->adap, p);
637 	} else if (opcode == CPL_ABORT_RPL_RSS) {
638 		const struct cpl_abort_rpl_rss *p = (void *)rsp;
639 
640 		hash_del_filter_rpl(q->adap, p);
641 	} else if (opcode == CPL_SRQ_TABLE_RPL) {
642 		const struct cpl_srq_table_rpl *p = (void *)rsp;
643 
644 		do_srq_table_rpl(q->adap, p);
645 	} else
646 		dev_err(q->adap->pdev_dev,
647 			"unexpected CPL %#x on FW event queue\n", opcode);
648 out:
649 	return 0;
650 }
651 
652 static void disable_msi(struct adapter *adapter)
653 {
654 	if (adapter->flags & CXGB4_USING_MSIX) {
655 		pci_disable_msix(adapter->pdev);
656 		adapter->flags &= ~CXGB4_USING_MSIX;
657 	} else if (adapter->flags & CXGB4_USING_MSI) {
658 		pci_disable_msi(adapter->pdev);
659 		adapter->flags &= ~CXGB4_USING_MSI;
660 	}
661 }
662 
663 /*
664  * Interrupt handler for non-data events used with MSI-X.
665  */
666 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
667 {
668 	struct adapter *adap = cookie;
669 	u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
670 
671 	if (v & PFSW_F) {
672 		adap->swintr = 1;
673 		t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
674 	}
675 	if (adap->flags & CXGB4_MASTER_PF)
676 		t4_slow_intr_handler(adap);
677 	return IRQ_HANDLED;
678 }
679 
680 /*
681  * Name the MSI-X interrupts.
682  */
683 static void name_msix_vecs(struct adapter *adap)
684 {
685 	int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
686 
687 	/* non-data interrupts */
688 	snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
689 
690 	/* FW events */
691 	snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
692 		 adap->port[0]->name);
693 
694 	/* Ethernet queues */
695 	for_each_port(adap, j) {
696 		struct net_device *d = adap->port[j];
697 		const struct port_info *pi = netdev_priv(d);
698 
699 		for (i = 0; i < pi->nqsets; i++, msi_idx++)
700 			snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
701 				 d->name, i);
702 	}
703 }
704 
705 static int request_msix_queue_irqs(struct adapter *adap)
706 {
707 	struct sge *s = &adap->sge;
708 	int err, ethqidx;
709 	int msi_index = 2;
710 
711 	err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
712 			  adap->msix_info[1].desc, &s->fw_evtq);
713 	if (err)
714 		return err;
715 
716 	for_each_ethrxq(s, ethqidx) {
717 		err = request_irq(adap->msix_info[msi_index].vec,
718 				  t4_sge_intr_msix, 0,
719 				  adap->msix_info[msi_index].desc,
720 				  &s->ethrxq[ethqidx].rspq);
721 		if (err)
722 			goto unwind;
723 		msi_index++;
724 	}
725 	return 0;
726 
727 unwind:
728 	while (--ethqidx >= 0)
729 		free_irq(adap->msix_info[--msi_index].vec,
730 			 &s->ethrxq[ethqidx].rspq);
731 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
732 	return err;
733 }
734 
735 static void free_msix_queue_irqs(struct adapter *adap)
736 {
737 	int i, msi_index = 2;
738 	struct sge *s = &adap->sge;
739 
740 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
741 	for_each_ethrxq(s, i)
742 		free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
743 }
744 
745 /**
746  *	cxgb4_write_rss - write the RSS table for a given port
747  *	@pi: the port
748  *	@queues: array of queue indices for RSS
749  *
750  *	Sets up the portion of the HW RSS table for the port's VI to distribute
751  *	packets to the Rx queues in @queues.
752  *	Should never be called before setting up sge eth rx queues
753  */
754 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
755 {
756 	u16 *rss;
757 	int i, err;
758 	struct adapter *adapter = pi->adapter;
759 	const struct sge_eth_rxq *rxq;
760 
761 	rxq = &adapter->sge.ethrxq[pi->first_qset];
762 	rss = kmalloc_array(pi->rss_size, sizeof(u16), GFP_KERNEL);
763 	if (!rss)
764 		return -ENOMEM;
765 
766 	/* map the queue indices to queue ids */
767 	for (i = 0; i < pi->rss_size; i++, queues++)
768 		rss[i] = rxq[*queues].rspq.abs_id;
769 
770 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
771 				  pi->rss_size, rss, pi->rss_size);
772 	/* If Tunnel All Lookup isn't specified in the global RSS
773 	 * Configuration, then we need to specify a default Ingress
774 	 * Queue for any ingress packets which aren't hashed.  We'll
775 	 * use our first ingress queue ...
776 	 */
777 	if (!err)
778 		err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
779 				       FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
780 				       FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
781 				       FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
782 				       FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
783 				       FW_RSS_VI_CONFIG_CMD_UDPEN_F,
784 				       rss[0]);
785 	kfree(rss);
786 	return err;
787 }
788 
789 /**
790  *	setup_rss - configure RSS
791  *	@adap: the adapter
792  *
793  *	Sets up RSS for each port.
794  */
795 static int setup_rss(struct adapter *adap)
796 {
797 	int i, j, err;
798 
799 	for_each_port(adap, i) {
800 		const struct port_info *pi = adap2pinfo(adap, i);
801 
802 		/* Fill default values with equal distribution */
803 		for (j = 0; j < pi->rss_size; j++)
804 			pi->rss[j] = j % pi->nqsets;
805 
806 		err = cxgb4_write_rss(pi, pi->rss);
807 		if (err)
808 			return err;
809 	}
810 	return 0;
811 }
812 
813 /*
814  * Return the channel of the ingress queue with the given qid.
815  */
816 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
817 {
818 	qid -= p->ingr_start;
819 	return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
820 }
821 
822 /*
823  * Wait until all NAPI handlers are descheduled.
824  */
825 static void quiesce_rx(struct adapter *adap)
826 {
827 	int i;
828 
829 	for (i = 0; i < adap->sge.ingr_sz; i++) {
830 		struct sge_rspq *q = adap->sge.ingr_map[i];
831 
832 		if (q && q->handler)
833 			napi_disable(&q->napi);
834 	}
835 }
836 
837 /* Disable interrupt and napi handler */
838 static void disable_interrupts(struct adapter *adap)
839 {
840 	if (adap->flags & CXGB4_FULL_INIT_DONE) {
841 		t4_intr_disable(adap);
842 		if (adap->flags & CXGB4_USING_MSIX) {
843 			free_msix_queue_irqs(adap);
844 			free_irq(adap->msix_info[0].vec, adap);
845 		} else {
846 			free_irq(adap->pdev->irq, adap);
847 		}
848 		quiesce_rx(adap);
849 	}
850 }
851 
852 /*
853  * Enable NAPI scheduling and interrupt generation for all Rx queues.
854  */
855 static void enable_rx(struct adapter *adap)
856 {
857 	int i;
858 
859 	for (i = 0; i < adap->sge.ingr_sz; i++) {
860 		struct sge_rspq *q = adap->sge.ingr_map[i];
861 
862 		if (!q)
863 			continue;
864 		if (q->handler)
865 			napi_enable(&q->napi);
866 
867 		/* 0-increment GTS to start the timer and enable interrupts */
868 		t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
869 			     SEINTARM_V(q->intr_params) |
870 			     INGRESSQID_V(q->cntxt_id));
871 	}
872 }
873 
874 
875 static int setup_fw_sge_queues(struct adapter *adap)
876 {
877 	struct sge *s = &adap->sge;
878 	int err = 0;
879 
880 	bitmap_zero(s->starving_fl, s->egr_sz);
881 	bitmap_zero(s->txq_maperr, s->egr_sz);
882 
883 	if (adap->flags & CXGB4_USING_MSIX)
884 		adap->msi_idx = 1;         /* vector 0 is for non-queue interrupts */
885 	else {
886 		err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
887 				       NULL, NULL, NULL, -1);
888 		if (err)
889 			return err;
890 		adap->msi_idx = -((int)s->intrq.abs_id + 1);
891 	}
892 
893 	err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
894 			       adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
895 	return err;
896 }
897 
898 /**
899  *	setup_sge_queues - configure SGE Tx/Rx/response queues
900  *	@adap: the adapter
901  *
902  *	Determines how many sets of SGE queues to use and initializes them.
903  *	We support multiple queue sets per port if we have MSI-X, otherwise
904  *	just one queue set per port.
905  */
906 static int setup_sge_queues(struct adapter *adap)
907 {
908 	int err, i, j;
909 	struct sge *s = &adap->sge;
910 	struct sge_uld_rxq_info *rxq_info = NULL;
911 	unsigned int cmplqid = 0;
912 
913 	if (is_uld(adap))
914 		rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
915 
916 	for_each_port(adap, i) {
917 		struct net_device *dev = adap->port[i];
918 		struct port_info *pi = netdev_priv(dev);
919 		struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
920 		struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
921 
922 		for (j = 0; j < pi->nqsets; j++, q++) {
923 			if (adap->msi_idx > 0)
924 				adap->msi_idx++;
925 			err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
926 					       adap->msi_idx, &q->fl,
927 					       t4_ethrx_handler,
928 					       NULL,
929 					       t4_get_tp_ch_map(adap,
930 								pi->tx_chan));
931 			if (err)
932 				goto freeout;
933 			q->rspq.idx = j;
934 			memset(&q->stats, 0, sizeof(q->stats));
935 		}
936 
937 		q = &s->ethrxq[pi->first_qset];
938 		for (j = 0; j < pi->nqsets; j++, t++, q++) {
939 			err = t4_sge_alloc_eth_txq(adap, t, dev,
940 					netdev_get_tx_queue(dev, j),
941 					q->rspq.cntxt_id,
942 					!!(adap->flags & CXGB4_SGE_DBQ_TIMER));
943 			if (err)
944 				goto freeout;
945 		}
946 	}
947 
948 	for_each_port(adap, i) {
949 		/* Note that cmplqid below is 0 if we don't
950 		 * have RDMA queues, and that's the right value.
951 		 */
952 		if (rxq_info)
953 			cmplqid	= rxq_info->uldrxq[i].rspq.cntxt_id;
954 
955 		err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
956 					    s->fw_evtq.cntxt_id, cmplqid);
957 		if (err)
958 			goto freeout;
959 	}
960 
961 	if (!is_t4(adap->params.chip)) {
962 		err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
963 					   netdev_get_tx_queue(adap->port[0], 0)
964 					   , s->fw_evtq.cntxt_id, false);
965 		if (err)
966 			goto freeout;
967 	}
968 
969 	t4_write_reg(adap, is_t4(adap->params.chip) ?
970 				MPS_TRC_RSS_CONTROL_A :
971 				MPS_T5_TRC_RSS_CONTROL_A,
972 		     RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
973 		     QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
974 	return 0;
975 freeout:
976 	dev_err(adap->pdev_dev, "Can't allocate queues, err=%d\n", -err);
977 	t4_free_sge_resources(adap);
978 	return err;
979 }
980 
981 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
982 			     struct net_device *sb_dev,
983 			     select_queue_fallback_t fallback)
984 {
985 	int txq;
986 
987 #ifdef CONFIG_CHELSIO_T4_DCB
988 	/* If a Data Center Bridging has been successfully negotiated on this
989 	 * link then we'll use the skb's priority to map it to a TX Queue.
990 	 * The skb's priority is determined via the VLAN Tag Priority Code
991 	 * Point field.
992 	 */
993 	if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
994 		u16 vlan_tci;
995 		int err;
996 
997 		err = vlan_get_tag(skb, &vlan_tci);
998 		if (unlikely(err)) {
999 			if (net_ratelimit())
1000 				netdev_warn(dev,
1001 					    "TX Packet without VLAN Tag on DCB Link\n");
1002 			txq = 0;
1003 		} else {
1004 			txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1005 #ifdef CONFIG_CHELSIO_T4_FCOE
1006 			if (skb->protocol == htons(ETH_P_FCOE))
1007 				txq = skb->priority & 0x7;
1008 #endif /* CONFIG_CHELSIO_T4_FCOE */
1009 		}
1010 		return txq;
1011 	}
1012 #endif /* CONFIG_CHELSIO_T4_DCB */
1013 
1014 	if (select_queue) {
1015 		txq = (skb_rx_queue_recorded(skb)
1016 			? skb_get_rx_queue(skb)
1017 			: smp_processor_id());
1018 
1019 		while (unlikely(txq >= dev->real_num_tx_queues))
1020 			txq -= dev->real_num_tx_queues;
1021 
1022 		return txq;
1023 	}
1024 
1025 	return fallback(dev, skb, NULL) % dev->real_num_tx_queues;
1026 }
1027 
1028 static int closest_timer(const struct sge *s, int time)
1029 {
1030 	int i, delta, match = 0, min_delta = INT_MAX;
1031 
1032 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1033 		delta = time - s->timer_val[i];
1034 		if (delta < 0)
1035 			delta = -delta;
1036 		if (delta < min_delta) {
1037 			min_delta = delta;
1038 			match = i;
1039 		}
1040 	}
1041 	return match;
1042 }
1043 
1044 static int closest_thres(const struct sge *s, int thres)
1045 {
1046 	int i, delta, match = 0, min_delta = INT_MAX;
1047 
1048 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1049 		delta = thres - s->counter_val[i];
1050 		if (delta < 0)
1051 			delta = -delta;
1052 		if (delta < min_delta) {
1053 			min_delta = delta;
1054 			match = i;
1055 		}
1056 	}
1057 	return match;
1058 }
1059 
1060 /**
1061  *	cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1062  *	@q: the Rx queue
1063  *	@us: the hold-off time in us, or 0 to disable timer
1064  *	@cnt: the hold-off packet count, or 0 to disable counter
1065  *
1066  *	Sets an Rx queue's interrupt hold-off time and packet count.  At least
1067  *	one of the two needs to be enabled for the queue to generate interrupts.
1068  */
1069 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1070 			       unsigned int us, unsigned int cnt)
1071 {
1072 	struct adapter *adap = q->adap;
1073 
1074 	if ((us | cnt) == 0)
1075 		cnt = 1;
1076 
1077 	if (cnt) {
1078 		int err;
1079 		u32 v, new_idx;
1080 
1081 		new_idx = closest_thres(&adap->sge, cnt);
1082 		if (q->desc && q->pktcnt_idx != new_idx) {
1083 			/* the queue has already been created, update it */
1084 			v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1085 			    FW_PARAMS_PARAM_X_V(
1086 					FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1087 			    FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1088 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1089 					    &v, &new_idx);
1090 			if (err)
1091 				return err;
1092 		}
1093 		q->pktcnt_idx = new_idx;
1094 	}
1095 
1096 	us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1097 	q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1098 	return 0;
1099 }
1100 
1101 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1102 {
1103 	const struct port_info *pi = netdev_priv(dev);
1104 	netdev_features_t changed = dev->features ^ features;
1105 	int err;
1106 
1107 	if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1108 		return 0;
1109 
1110 	err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1111 			    -1, -1, -1,
1112 			    !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1113 	if (unlikely(err))
1114 		dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1115 	return err;
1116 }
1117 
1118 static int setup_debugfs(struct adapter *adap)
1119 {
1120 	if (IS_ERR_OR_NULL(adap->debugfs_root))
1121 		return -1;
1122 
1123 #ifdef CONFIG_DEBUG_FS
1124 	t4_setup_debugfs(adap);
1125 #endif
1126 	return 0;
1127 }
1128 
1129 /*
1130  * upper-layer driver support
1131  */
1132 
1133 /*
1134  * Allocate an active-open TID and set it to the supplied value.
1135  */
1136 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1137 {
1138 	int atid = -1;
1139 
1140 	spin_lock_bh(&t->atid_lock);
1141 	if (t->afree) {
1142 		union aopen_entry *p = t->afree;
1143 
1144 		atid = (p - t->atid_tab) + t->atid_base;
1145 		t->afree = p->next;
1146 		p->data = data;
1147 		t->atids_in_use++;
1148 	}
1149 	spin_unlock_bh(&t->atid_lock);
1150 	return atid;
1151 }
1152 EXPORT_SYMBOL(cxgb4_alloc_atid);
1153 
1154 /*
1155  * Release an active-open TID.
1156  */
1157 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1158 {
1159 	union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1160 
1161 	spin_lock_bh(&t->atid_lock);
1162 	p->next = t->afree;
1163 	t->afree = p;
1164 	t->atids_in_use--;
1165 	spin_unlock_bh(&t->atid_lock);
1166 }
1167 EXPORT_SYMBOL(cxgb4_free_atid);
1168 
1169 /*
1170  * Allocate a server TID and set it to the supplied value.
1171  */
1172 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1173 {
1174 	int stid;
1175 
1176 	spin_lock_bh(&t->stid_lock);
1177 	if (family == PF_INET) {
1178 		stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1179 		if (stid < t->nstids)
1180 			__set_bit(stid, t->stid_bmap);
1181 		else
1182 			stid = -1;
1183 	} else {
1184 		stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1185 		if (stid < 0)
1186 			stid = -1;
1187 	}
1188 	if (stid >= 0) {
1189 		t->stid_tab[stid].data = data;
1190 		stid += t->stid_base;
1191 		/* IPv6 requires max of 520 bits or 16 cells in TCAM
1192 		 * This is equivalent to 4 TIDs. With CLIP enabled it
1193 		 * needs 2 TIDs.
1194 		 */
1195 		if (family == PF_INET6) {
1196 			t->stids_in_use += 2;
1197 			t->v6_stids_in_use += 2;
1198 		} else {
1199 			t->stids_in_use++;
1200 		}
1201 	}
1202 	spin_unlock_bh(&t->stid_lock);
1203 	return stid;
1204 }
1205 EXPORT_SYMBOL(cxgb4_alloc_stid);
1206 
1207 /* Allocate a server filter TID and set it to the supplied value.
1208  */
1209 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1210 {
1211 	int stid;
1212 
1213 	spin_lock_bh(&t->stid_lock);
1214 	if (family == PF_INET) {
1215 		stid = find_next_zero_bit(t->stid_bmap,
1216 				t->nstids + t->nsftids, t->nstids);
1217 		if (stid < (t->nstids + t->nsftids))
1218 			__set_bit(stid, t->stid_bmap);
1219 		else
1220 			stid = -1;
1221 	} else {
1222 		stid = -1;
1223 	}
1224 	if (stid >= 0) {
1225 		t->stid_tab[stid].data = data;
1226 		stid -= t->nstids;
1227 		stid += t->sftid_base;
1228 		t->sftids_in_use++;
1229 	}
1230 	spin_unlock_bh(&t->stid_lock);
1231 	return stid;
1232 }
1233 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1234 
1235 /* Release a server TID.
1236  */
1237 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1238 {
1239 	/* Is it a server filter TID? */
1240 	if (t->nsftids && (stid >= t->sftid_base)) {
1241 		stid -= t->sftid_base;
1242 		stid += t->nstids;
1243 	} else {
1244 		stid -= t->stid_base;
1245 	}
1246 
1247 	spin_lock_bh(&t->stid_lock);
1248 	if (family == PF_INET)
1249 		__clear_bit(stid, t->stid_bmap);
1250 	else
1251 		bitmap_release_region(t->stid_bmap, stid, 1);
1252 	t->stid_tab[stid].data = NULL;
1253 	if (stid < t->nstids) {
1254 		if (family == PF_INET6) {
1255 			t->stids_in_use -= 2;
1256 			t->v6_stids_in_use -= 2;
1257 		} else {
1258 			t->stids_in_use--;
1259 		}
1260 	} else {
1261 		t->sftids_in_use--;
1262 	}
1263 
1264 	spin_unlock_bh(&t->stid_lock);
1265 }
1266 EXPORT_SYMBOL(cxgb4_free_stid);
1267 
1268 /*
1269  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1270  */
1271 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1272 			   unsigned int tid)
1273 {
1274 	struct cpl_tid_release *req;
1275 
1276 	set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1277 	req = __skb_put(skb, sizeof(*req));
1278 	INIT_TP_WR(req, tid);
1279 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1280 }
1281 
1282 /*
1283  * Queue a TID release request and if necessary schedule a work queue to
1284  * process it.
1285  */
1286 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1287 				    unsigned int tid)
1288 {
1289 	void **p = &t->tid_tab[tid];
1290 	struct adapter *adap = container_of(t, struct adapter, tids);
1291 
1292 	spin_lock_bh(&adap->tid_release_lock);
1293 	*p = adap->tid_release_head;
1294 	/* Low 2 bits encode the Tx channel number */
1295 	adap->tid_release_head = (void **)((uintptr_t)p | chan);
1296 	if (!adap->tid_release_task_busy) {
1297 		adap->tid_release_task_busy = true;
1298 		queue_work(adap->workq, &adap->tid_release_task);
1299 	}
1300 	spin_unlock_bh(&adap->tid_release_lock);
1301 }
1302 
1303 /*
1304  * Process the list of pending TID release requests.
1305  */
1306 static void process_tid_release_list(struct work_struct *work)
1307 {
1308 	struct sk_buff *skb;
1309 	struct adapter *adap;
1310 
1311 	adap = container_of(work, struct adapter, tid_release_task);
1312 
1313 	spin_lock_bh(&adap->tid_release_lock);
1314 	while (adap->tid_release_head) {
1315 		void **p = adap->tid_release_head;
1316 		unsigned int chan = (uintptr_t)p & 3;
1317 		p = (void *)p - chan;
1318 
1319 		adap->tid_release_head = *p;
1320 		*p = NULL;
1321 		spin_unlock_bh(&adap->tid_release_lock);
1322 
1323 		while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1324 					 GFP_KERNEL)))
1325 			schedule_timeout_uninterruptible(1);
1326 
1327 		mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1328 		t4_ofld_send(adap, skb);
1329 		spin_lock_bh(&adap->tid_release_lock);
1330 	}
1331 	adap->tid_release_task_busy = false;
1332 	spin_unlock_bh(&adap->tid_release_lock);
1333 }
1334 
1335 /*
1336  * Release a TID and inform HW.  If we are unable to allocate the release
1337  * message we defer to a work queue.
1338  */
1339 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1340 		      unsigned short family)
1341 {
1342 	struct sk_buff *skb;
1343 	struct adapter *adap = container_of(t, struct adapter, tids);
1344 
1345 	WARN_ON(tid >= t->ntids);
1346 
1347 	if (t->tid_tab[tid]) {
1348 		t->tid_tab[tid] = NULL;
1349 		atomic_dec(&t->conns_in_use);
1350 		if (t->hash_base && (tid >= t->hash_base)) {
1351 			if (family == AF_INET6)
1352 				atomic_sub(2, &t->hash_tids_in_use);
1353 			else
1354 				atomic_dec(&t->hash_tids_in_use);
1355 		} else {
1356 			if (family == AF_INET6)
1357 				atomic_sub(2, &t->tids_in_use);
1358 			else
1359 				atomic_dec(&t->tids_in_use);
1360 		}
1361 	}
1362 
1363 	skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1364 	if (likely(skb)) {
1365 		mk_tid_release(skb, chan, tid);
1366 		t4_ofld_send(adap, skb);
1367 	} else
1368 		cxgb4_queue_tid_release(t, chan, tid);
1369 }
1370 EXPORT_SYMBOL(cxgb4_remove_tid);
1371 
1372 /*
1373  * Allocate and initialize the TID tables.  Returns 0 on success.
1374  */
1375 static int tid_init(struct tid_info *t)
1376 {
1377 	struct adapter *adap = container_of(t, struct adapter, tids);
1378 	unsigned int max_ftids = t->nftids + t->nsftids;
1379 	unsigned int natids = t->natids;
1380 	unsigned int stid_bmap_size;
1381 	unsigned int ftid_bmap_size;
1382 	size_t size;
1383 
1384 	stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1385 	ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1386 	size = t->ntids * sizeof(*t->tid_tab) +
1387 	       natids * sizeof(*t->atid_tab) +
1388 	       t->nstids * sizeof(*t->stid_tab) +
1389 	       t->nsftids * sizeof(*t->stid_tab) +
1390 	       stid_bmap_size * sizeof(long) +
1391 	       max_ftids * sizeof(*t->ftid_tab) +
1392 	       ftid_bmap_size * sizeof(long);
1393 
1394 	t->tid_tab = kvzalloc(size, GFP_KERNEL);
1395 	if (!t->tid_tab)
1396 		return -ENOMEM;
1397 
1398 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1399 	t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1400 	t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1401 	t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1402 	t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1403 	spin_lock_init(&t->stid_lock);
1404 	spin_lock_init(&t->atid_lock);
1405 	spin_lock_init(&t->ftid_lock);
1406 
1407 	t->stids_in_use = 0;
1408 	t->v6_stids_in_use = 0;
1409 	t->sftids_in_use = 0;
1410 	t->afree = NULL;
1411 	t->atids_in_use = 0;
1412 	atomic_set(&t->tids_in_use, 0);
1413 	atomic_set(&t->conns_in_use, 0);
1414 	atomic_set(&t->hash_tids_in_use, 0);
1415 
1416 	/* Setup the free list for atid_tab and clear the stid bitmap. */
1417 	if (natids) {
1418 		while (--natids)
1419 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1420 		t->afree = t->atid_tab;
1421 	}
1422 
1423 	if (is_offload(adap)) {
1424 		bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1425 		/* Reserve stid 0 for T4/T5 adapters */
1426 		if (!t->stid_base &&
1427 		    CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1428 			__set_bit(0, t->stid_bmap);
1429 	}
1430 
1431 	bitmap_zero(t->ftid_bmap, t->nftids);
1432 	return 0;
1433 }
1434 
1435 /**
1436  *	cxgb4_create_server - create an IP server
1437  *	@dev: the device
1438  *	@stid: the server TID
1439  *	@sip: local IP address to bind server to
1440  *	@sport: the server's TCP port
1441  *	@queue: queue to direct messages from this server to
1442  *
1443  *	Create an IP server for the given port and address.
1444  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
1445  */
1446 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1447 			__be32 sip, __be16 sport, __be16 vlan,
1448 			unsigned int queue)
1449 {
1450 	unsigned int chan;
1451 	struct sk_buff *skb;
1452 	struct adapter *adap;
1453 	struct cpl_pass_open_req *req;
1454 	int ret;
1455 
1456 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1457 	if (!skb)
1458 		return -ENOMEM;
1459 
1460 	adap = netdev2adap(dev);
1461 	req = __skb_put(skb, sizeof(*req));
1462 	INIT_TP_WR(req, 0);
1463 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1464 	req->local_port = sport;
1465 	req->peer_port = htons(0);
1466 	req->local_ip = sip;
1467 	req->peer_ip = htonl(0);
1468 	chan = rxq_to_chan(&adap->sge, queue);
1469 	req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1470 	req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1471 				SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1472 	ret = t4_mgmt_tx(adap, skb);
1473 	return net_xmit_eval(ret);
1474 }
1475 EXPORT_SYMBOL(cxgb4_create_server);
1476 
1477 /*	cxgb4_create_server6 - create an IPv6 server
1478  *	@dev: the device
1479  *	@stid: the server TID
1480  *	@sip: local IPv6 address to bind server to
1481  *	@sport: the server's TCP port
1482  *	@queue: queue to direct messages from this server to
1483  *
1484  *	Create an IPv6 server for the given port and address.
1485  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
1486  */
1487 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1488 			 const struct in6_addr *sip, __be16 sport,
1489 			 unsigned int queue)
1490 {
1491 	unsigned int chan;
1492 	struct sk_buff *skb;
1493 	struct adapter *adap;
1494 	struct cpl_pass_open_req6 *req;
1495 	int ret;
1496 
1497 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1498 	if (!skb)
1499 		return -ENOMEM;
1500 
1501 	adap = netdev2adap(dev);
1502 	req = __skb_put(skb, sizeof(*req));
1503 	INIT_TP_WR(req, 0);
1504 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1505 	req->local_port = sport;
1506 	req->peer_port = htons(0);
1507 	req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1508 	req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1509 	req->peer_ip_hi = cpu_to_be64(0);
1510 	req->peer_ip_lo = cpu_to_be64(0);
1511 	chan = rxq_to_chan(&adap->sge, queue);
1512 	req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1513 	req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1514 				SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1515 	ret = t4_mgmt_tx(adap, skb);
1516 	return net_xmit_eval(ret);
1517 }
1518 EXPORT_SYMBOL(cxgb4_create_server6);
1519 
1520 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1521 			unsigned int queue, bool ipv6)
1522 {
1523 	struct sk_buff *skb;
1524 	struct adapter *adap;
1525 	struct cpl_close_listsvr_req *req;
1526 	int ret;
1527 
1528 	adap = netdev2adap(dev);
1529 
1530 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1531 	if (!skb)
1532 		return -ENOMEM;
1533 
1534 	req = __skb_put(skb, sizeof(*req));
1535 	INIT_TP_WR(req, 0);
1536 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1537 	req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1538 				LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1539 	ret = t4_mgmt_tx(adap, skb);
1540 	return net_xmit_eval(ret);
1541 }
1542 EXPORT_SYMBOL(cxgb4_remove_server);
1543 
1544 /**
1545  *	cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1546  *	@mtus: the HW MTU table
1547  *	@mtu: the target MTU
1548  *	@idx: index of selected entry in the MTU table
1549  *
1550  *	Returns the index and the value in the HW MTU table that is closest to
1551  *	but does not exceed @mtu, unless @mtu is smaller than any value in the
1552  *	table, in which case that smallest available value is selected.
1553  */
1554 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1555 			    unsigned int *idx)
1556 {
1557 	unsigned int i = 0;
1558 
1559 	while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1560 		++i;
1561 	if (idx)
1562 		*idx = i;
1563 	return mtus[i];
1564 }
1565 EXPORT_SYMBOL(cxgb4_best_mtu);
1566 
1567 /**
1568  *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1569  *     @mtus: the HW MTU table
1570  *     @header_size: Header Size
1571  *     @data_size_max: maximum Data Segment Size
1572  *     @data_size_align: desired Data Segment Size Alignment (2^N)
1573  *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1574  *
1575  *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
1576  *     MTU Table based solely on a Maximum MTU parameter, we break that
1577  *     parameter up into a Header Size and Maximum Data Segment Size, and
1578  *     provide a desired Data Segment Size Alignment.  If we find an MTU in
1579  *     the Hardware MTU Table which will result in a Data Segment Size with
1580  *     the requested alignment _and_ that MTU isn't "too far" from the
1581  *     closest MTU, then we'll return that rather than the closest MTU.
1582  */
1583 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1584 				    unsigned short header_size,
1585 				    unsigned short data_size_max,
1586 				    unsigned short data_size_align,
1587 				    unsigned int *mtu_idxp)
1588 {
1589 	unsigned short max_mtu = header_size + data_size_max;
1590 	unsigned short data_size_align_mask = data_size_align - 1;
1591 	int mtu_idx, aligned_mtu_idx;
1592 
1593 	/* Scan the MTU Table till we find an MTU which is larger than our
1594 	 * Maximum MTU or we reach the end of the table.  Along the way,
1595 	 * record the last MTU found, if any, which will result in a Data
1596 	 * Segment Length matching the requested alignment.
1597 	 */
1598 	for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1599 		unsigned short data_size = mtus[mtu_idx] - header_size;
1600 
1601 		/* If this MTU minus the Header Size would result in a
1602 		 * Data Segment Size of the desired alignment, remember it.
1603 		 */
1604 		if ((data_size & data_size_align_mask) == 0)
1605 			aligned_mtu_idx = mtu_idx;
1606 
1607 		/* If we're not at the end of the Hardware MTU Table and the
1608 		 * next element is larger than our Maximum MTU, drop out of
1609 		 * the loop.
1610 		 */
1611 		if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1612 			break;
1613 	}
1614 
1615 	/* If we fell out of the loop because we ran to the end of the table,
1616 	 * then we just have to use the last [largest] entry.
1617 	 */
1618 	if (mtu_idx == NMTUS)
1619 		mtu_idx--;
1620 
1621 	/* If we found an MTU which resulted in the requested Data Segment
1622 	 * Length alignment and that's "not far" from the largest MTU which is
1623 	 * less than or equal to the maximum MTU, then use that.
1624 	 */
1625 	if (aligned_mtu_idx >= 0 &&
1626 	    mtu_idx - aligned_mtu_idx <= 1)
1627 		mtu_idx = aligned_mtu_idx;
1628 
1629 	/* If the caller has passed in an MTU Index pointer, pass the
1630 	 * MTU Index back.  Return the MTU value.
1631 	 */
1632 	if (mtu_idxp)
1633 		*mtu_idxp = mtu_idx;
1634 	return mtus[mtu_idx];
1635 }
1636 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1637 
1638 /**
1639  *	cxgb4_port_chan - get the HW channel of a port
1640  *	@dev: the net device for the port
1641  *
1642  *	Return the HW Tx channel of the given port.
1643  */
1644 unsigned int cxgb4_port_chan(const struct net_device *dev)
1645 {
1646 	return netdev2pinfo(dev)->tx_chan;
1647 }
1648 EXPORT_SYMBOL(cxgb4_port_chan);
1649 
1650 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1651 {
1652 	struct adapter *adap = netdev2adap(dev);
1653 	u32 v1, v2, lp_count, hp_count;
1654 
1655 	v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1656 	v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1657 	if (is_t4(adap->params.chip)) {
1658 		lp_count = LP_COUNT_G(v1);
1659 		hp_count = HP_COUNT_G(v1);
1660 	} else {
1661 		lp_count = LP_COUNT_T5_G(v1);
1662 		hp_count = HP_COUNT_T5_G(v2);
1663 	}
1664 	return lpfifo ? lp_count : hp_count;
1665 }
1666 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1667 
1668 /**
1669  *	cxgb4_port_viid - get the VI id of a port
1670  *	@dev: the net device for the port
1671  *
1672  *	Return the VI id of the given port.
1673  */
1674 unsigned int cxgb4_port_viid(const struct net_device *dev)
1675 {
1676 	return netdev2pinfo(dev)->viid;
1677 }
1678 EXPORT_SYMBOL(cxgb4_port_viid);
1679 
1680 /**
1681  *	cxgb4_port_idx - get the index of a port
1682  *	@dev: the net device for the port
1683  *
1684  *	Return the index of the given port.
1685  */
1686 unsigned int cxgb4_port_idx(const struct net_device *dev)
1687 {
1688 	return netdev2pinfo(dev)->port_id;
1689 }
1690 EXPORT_SYMBOL(cxgb4_port_idx);
1691 
1692 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1693 			 struct tp_tcp_stats *v6)
1694 {
1695 	struct adapter *adap = pci_get_drvdata(pdev);
1696 
1697 	spin_lock(&adap->stats_lock);
1698 	t4_tp_get_tcp_stats(adap, v4, v6, false);
1699 	spin_unlock(&adap->stats_lock);
1700 }
1701 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1702 
1703 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1704 		      const unsigned int *pgsz_order)
1705 {
1706 	struct adapter *adap = netdev2adap(dev);
1707 
1708 	t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1709 	t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1710 		     HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1711 		     HPZ3_V(pgsz_order[3]));
1712 }
1713 EXPORT_SYMBOL(cxgb4_iscsi_init);
1714 
1715 int cxgb4_flush_eq_cache(struct net_device *dev)
1716 {
1717 	struct adapter *adap = netdev2adap(dev);
1718 
1719 	return t4_sge_ctxt_flush(adap, adap->mbox, CTXT_EGRESS);
1720 }
1721 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1722 
1723 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1724 {
1725 	u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1726 	__be64 indices;
1727 	int ret;
1728 
1729 	spin_lock(&adap->win0_lock);
1730 	ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1731 			   sizeof(indices), (__be32 *)&indices,
1732 			   T4_MEMORY_READ);
1733 	spin_unlock(&adap->win0_lock);
1734 	if (!ret) {
1735 		*cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1736 		*pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1737 	}
1738 	return ret;
1739 }
1740 
1741 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1742 			u16 size)
1743 {
1744 	struct adapter *adap = netdev2adap(dev);
1745 	u16 hw_pidx, hw_cidx;
1746 	int ret;
1747 
1748 	ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1749 	if (ret)
1750 		goto out;
1751 
1752 	if (pidx != hw_pidx) {
1753 		u16 delta;
1754 		u32 val;
1755 
1756 		if (pidx >= hw_pidx)
1757 			delta = pidx - hw_pidx;
1758 		else
1759 			delta = size - hw_pidx + pidx;
1760 
1761 		if (is_t4(adap->params.chip))
1762 			val = PIDX_V(delta);
1763 		else
1764 			val = PIDX_T5_V(delta);
1765 		wmb();
1766 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1767 			     QID_V(qid) | val);
1768 	}
1769 out:
1770 	return ret;
1771 }
1772 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1773 
1774 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1775 {
1776 	u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1777 	u32 edc0_end, edc1_end, mc0_end, mc1_end;
1778 	u32 offset, memtype, memaddr;
1779 	struct adapter *adap;
1780 	u32 hma_size = 0;
1781 	int ret;
1782 
1783 	adap = netdev2adap(dev);
1784 
1785 	offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1786 
1787 	/* Figure out where the offset lands in the Memory Type/Address scheme.
1788 	 * This code assumes that the memory is laid out starting at offset 0
1789 	 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1790 	 * and EDC1.  Some cards will have neither MC0 nor MC1, most cards have
1791 	 * MC0, and some have both MC0 and MC1.
1792 	 */
1793 	size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1794 	edc0_size = EDRAM0_SIZE_G(size) << 20;
1795 	size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1796 	edc1_size = EDRAM1_SIZE_G(size) << 20;
1797 	size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1798 	mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1799 
1800 	if (t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A) & HMA_MUX_F) {
1801 		size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1802 		hma_size = EXT_MEM1_SIZE_G(size) << 20;
1803 	}
1804 	edc0_end = edc0_size;
1805 	edc1_end = edc0_end + edc1_size;
1806 	mc0_end = edc1_end + mc0_size;
1807 
1808 	if (offset < edc0_end) {
1809 		memtype = MEM_EDC0;
1810 		memaddr = offset;
1811 	} else if (offset < edc1_end) {
1812 		memtype = MEM_EDC1;
1813 		memaddr = offset - edc0_end;
1814 	} else {
1815 		if (hma_size && (offset < (edc1_end + hma_size))) {
1816 			memtype = MEM_HMA;
1817 			memaddr = offset - edc1_end;
1818 		} else if (offset < mc0_end) {
1819 			memtype = MEM_MC0;
1820 			memaddr = offset - edc1_end;
1821 		} else if (is_t5(adap->params.chip)) {
1822 			size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1823 			mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1824 			mc1_end = mc0_end + mc1_size;
1825 			if (offset < mc1_end) {
1826 				memtype = MEM_MC1;
1827 				memaddr = offset - mc0_end;
1828 			} else {
1829 				/* offset beyond the end of any memory */
1830 				goto err;
1831 			}
1832 		} else {
1833 			/* T4/T6 only has a single memory channel */
1834 			goto err;
1835 		}
1836 	}
1837 
1838 	spin_lock(&adap->win0_lock);
1839 	ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1840 	spin_unlock(&adap->win0_lock);
1841 	return ret;
1842 
1843 err:
1844 	dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1845 		stag, offset);
1846 	return -EINVAL;
1847 }
1848 EXPORT_SYMBOL(cxgb4_read_tpte);
1849 
1850 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1851 {
1852 	u32 hi, lo;
1853 	struct adapter *adap;
1854 
1855 	adap = netdev2adap(dev);
1856 	lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1857 	hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1858 
1859 	return ((u64)hi << 32) | (u64)lo;
1860 }
1861 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1862 
1863 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1864 			 unsigned int qid,
1865 			 enum cxgb4_bar2_qtype qtype,
1866 			 int user,
1867 			 u64 *pbar2_qoffset,
1868 			 unsigned int *pbar2_qid)
1869 {
1870 	return t4_bar2_sge_qregs(netdev2adap(dev),
1871 				 qid,
1872 				 (qtype == CXGB4_BAR2_QTYPE_EGRESS
1873 				  ? T4_BAR2_QTYPE_EGRESS
1874 				  : T4_BAR2_QTYPE_INGRESS),
1875 				 user,
1876 				 pbar2_qoffset,
1877 				 pbar2_qid);
1878 }
1879 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1880 
1881 static struct pci_driver cxgb4_driver;
1882 
1883 static void check_neigh_update(struct neighbour *neigh)
1884 {
1885 	const struct device *parent;
1886 	const struct net_device *netdev = neigh->dev;
1887 
1888 	if (is_vlan_dev(netdev))
1889 		netdev = vlan_dev_real_dev(netdev);
1890 	parent = netdev->dev.parent;
1891 	if (parent && parent->driver == &cxgb4_driver.driver)
1892 		t4_l2t_update(dev_get_drvdata(parent), neigh);
1893 }
1894 
1895 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1896 		       void *data)
1897 {
1898 	switch (event) {
1899 	case NETEVENT_NEIGH_UPDATE:
1900 		check_neigh_update(data);
1901 		break;
1902 	case NETEVENT_REDIRECT:
1903 	default:
1904 		break;
1905 	}
1906 	return 0;
1907 }
1908 
1909 static bool netevent_registered;
1910 static struct notifier_block cxgb4_netevent_nb = {
1911 	.notifier_call = netevent_cb
1912 };
1913 
1914 static void drain_db_fifo(struct adapter *adap, int usecs)
1915 {
1916 	u32 v1, v2, lp_count, hp_count;
1917 
1918 	do {
1919 		v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1920 		v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1921 		if (is_t4(adap->params.chip)) {
1922 			lp_count = LP_COUNT_G(v1);
1923 			hp_count = HP_COUNT_G(v1);
1924 		} else {
1925 			lp_count = LP_COUNT_T5_G(v1);
1926 			hp_count = HP_COUNT_T5_G(v2);
1927 		}
1928 
1929 		if (lp_count == 0 && hp_count == 0)
1930 			break;
1931 		set_current_state(TASK_UNINTERRUPTIBLE);
1932 		schedule_timeout(usecs_to_jiffies(usecs));
1933 	} while (1);
1934 }
1935 
1936 static void disable_txq_db(struct sge_txq *q)
1937 {
1938 	unsigned long flags;
1939 
1940 	spin_lock_irqsave(&q->db_lock, flags);
1941 	q->db_disabled = 1;
1942 	spin_unlock_irqrestore(&q->db_lock, flags);
1943 }
1944 
1945 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1946 {
1947 	spin_lock_irq(&q->db_lock);
1948 	if (q->db_pidx_inc) {
1949 		/* Make sure that all writes to the TX descriptors
1950 		 * are committed before we tell HW about them.
1951 		 */
1952 		wmb();
1953 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1954 			     QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1955 		q->db_pidx_inc = 0;
1956 	}
1957 	q->db_disabled = 0;
1958 	spin_unlock_irq(&q->db_lock);
1959 }
1960 
1961 static void disable_dbs(struct adapter *adap)
1962 {
1963 	int i;
1964 
1965 	for_each_ethrxq(&adap->sge, i)
1966 		disable_txq_db(&adap->sge.ethtxq[i].q);
1967 	if (is_offload(adap)) {
1968 		struct sge_uld_txq_info *txq_info =
1969 			adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1970 
1971 		if (txq_info) {
1972 			for_each_ofldtxq(&adap->sge, i) {
1973 				struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1974 
1975 				disable_txq_db(&txq->q);
1976 			}
1977 		}
1978 	}
1979 	for_each_port(adap, i)
1980 		disable_txq_db(&adap->sge.ctrlq[i].q);
1981 }
1982 
1983 static void enable_dbs(struct adapter *adap)
1984 {
1985 	int i;
1986 
1987 	for_each_ethrxq(&adap->sge, i)
1988 		enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1989 	if (is_offload(adap)) {
1990 		struct sge_uld_txq_info *txq_info =
1991 			adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1992 
1993 		if (txq_info) {
1994 			for_each_ofldtxq(&adap->sge, i) {
1995 				struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1996 
1997 				enable_txq_db(adap, &txq->q);
1998 			}
1999 		}
2000 	}
2001 	for_each_port(adap, i)
2002 		enable_txq_db(adap, &adap->sge.ctrlq[i].q);
2003 }
2004 
2005 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
2006 {
2007 	enum cxgb4_uld type = CXGB4_ULD_RDMA;
2008 
2009 	if (adap->uld && adap->uld[type].handle)
2010 		adap->uld[type].control(adap->uld[type].handle, cmd);
2011 }
2012 
2013 static void process_db_full(struct work_struct *work)
2014 {
2015 	struct adapter *adap;
2016 
2017 	adap = container_of(work, struct adapter, db_full_task);
2018 
2019 	drain_db_fifo(adap, dbfifo_drain_delay);
2020 	enable_dbs(adap);
2021 	notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2022 	if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2023 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2024 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
2025 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
2026 	else
2027 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2028 				 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
2029 }
2030 
2031 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2032 {
2033 	u16 hw_pidx, hw_cidx;
2034 	int ret;
2035 
2036 	spin_lock_irq(&q->db_lock);
2037 	ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2038 	if (ret)
2039 		goto out;
2040 	if (q->db_pidx != hw_pidx) {
2041 		u16 delta;
2042 		u32 val;
2043 
2044 		if (q->db_pidx >= hw_pidx)
2045 			delta = q->db_pidx - hw_pidx;
2046 		else
2047 			delta = q->size - hw_pidx + q->db_pidx;
2048 
2049 		if (is_t4(adap->params.chip))
2050 			val = PIDX_V(delta);
2051 		else
2052 			val = PIDX_T5_V(delta);
2053 		wmb();
2054 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2055 			     QID_V(q->cntxt_id) | val);
2056 	}
2057 out:
2058 	q->db_disabled = 0;
2059 	q->db_pidx_inc = 0;
2060 	spin_unlock_irq(&q->db_lock);
2061 	if (ret)
2062 		CH_WARN(adap, "DB drop recovery failed.\n");
2063 }
2064 
2065 static void recover_all_queues(struct adapter *adap)
2066 {
2067 	int i;
2068 
2069 	for_each_ethrxq(&adap->sge, i)
2070 		sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2071 	if (is_offload(adap)) {
2072 		struct sge_uld_txq_info *txq_info =
2073 			adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2074 		if (txq_info) {
2075 			for_each_ofldtxq(&adap->sge, i) {
2076 				struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2077 
2078 				sync_txq_pidx(adap, &txq->q);
2079 			}
2080 		}
2081 	}
2082 	for_each_port(adap, i)
2083 		sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2084 }
2085 
2086 static void process_db_drop(struct work_struct *work)
2087 {
2088 	struct adapter *adap;
2089 
2090 	adap = container_of(work, struct adapter, db_drop_task);
2091 
2092 	if (is_t4(adap->params.chip)) {
2093 		drain_db_fifo(adap, dbfifo_drain_delay);
2094 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2095 		drain_db_fifo(adap, dbfifo_drain_delay);
2096 		recover_all_queues(adap);
2097 		drain_db_fifo(adap, dbfifo_drain_delay);
2098 		enable_dbs(adap);
2099 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2100 	} else if (is_t5(adap->params.chip)) {
2101 		u32 dropped_db = t4_read_reg(adap, 0x010ac);
2102 		u16 qid = (dropped_db >> 15) & 0x1ffff;
2103 		u16 pidx_inc = dropped_db & 0x1fff;
2104 		u64 bar2_qoffset;
2105 		unsigned int bar2_qid;
2106 		int ret;
2107 
2108 		ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2109 					0, &bar2_qoffset, &bar2_qid);
2110 		if (ret)
2111 			dev_err(adap->pdev_dev, "doorbell drop recovery: "
2112 				"qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2113 		else
2114 			writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2115 			       adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2116 
2117 		/* Re-enable BAR2 WC */
2118 		t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2119 	}
2120 
2121 	if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2122 		t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2123 }
2124 
2125 void t4_db_full(struct adapter *adap)
2126 {
2127 	if (is_t4(adap->params.chip)) {
2128 		disable_dbs(adap);
2129 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2130 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2131 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2132 		queue_work(adap->workq, &adap->db_full_task);
2133 	}
2134 }
2135 
2136 void t4_db_dropped(struct adapter *adap)
2137 {
2138 	if (is_t4(adap->params.chip)) {
2139 		disable_dbs(adap);
2140 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2141 	}
2142 	queue_work(adap->workq, &adap->db_drop_task);
2143 }
2144 
2145 void t4_register_netevent_notifier(void)
2146 {
2147 	if (!netevent_registered) {
2148 		register_netevent_notifier(&cxgb4_netevent_nb);
2149 		netevent_registered = true;
2150 	}
2151 }
2152 
2153 static void detach_ulds(struct adapter *adap)
2154 {
2155 	unsigned int i;
2156 
2157 	mutex_lock(&uld_mutex);
2158 	list_del(&adap->list_node);
2159 
2160 	for (i = 0; i < CXGB4_ULD_MAX; i++)
2161 		if (adap->uld && adap->uld[i].handle)
2162 			adap->uld[i].state_change(adap->uld[i].handle,
2163 					     CXGB4_STATE_DETACH);
2164 
2165 	if (netevent_registered && list_empty(&adapter_list)) {
2166 		unregister_netevent_notifier(&cxgb4_netevent_nb);
2167 		netevent_registered = false;
2168 	}
2169 	mutex_unlock(&uld_mutex);
2170 }
2171 
2172 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2173 {
2174 	unsigned int i;
2175 
2176 	mutex_lock(&uld_mutex);
2177 	for (i = 0; i < CXGB4_ULD_MAX; i++)
2178 		if (adap->uld && adap->uld[i].handle)
2179 			adap->uld[i].state_change(adap->uld[i].handle,
2180 						  new_state);
2181 	mutex_unlock(&uld_mutex);
2182 }
2183 
2184 #if IS_ENABLED(CONFIG_IPV6)
2185 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2186 				   unsigned long event, void *data)
2187 {
2188 	struct inet6_ifaddr *ifa = data;
2189 	struct net_device *event_dev = ifa->idev->dev;
2190 	const struct device *parent = NULL;
2191 #if IS_ENABLED(CONFIG_BONDING)
2192 	struct adapter *adap;
2193 #endif
2194 	if (is_vlan_dev(event_dev))
2195 		event_dev = vlan_dev_real_dev(event_dev);
2196 #if IS_ENABLED(CONFIG_BONDING)
2197 	if (event_dev->flags & IFF_MASTER) {
2198 		list_for_each_entry(adap, &adapter_list, list_node) {
2199 			switch (event) {
2200 			case NETDEV_UP:
2201 				cxgb4_clip_get(adap->port[0],
2202 					       (const u32 *)ifa, 1);
2203 				break;
2204 			case NETDEV_DOWN:
2205 				cxgb4_clip_release(adap->port[0],
2206 						   (const u32 *)ifa, 1);
2207 				break;
2208 			default:
2209 				break;
2210 			}
2211 		}
2212 		return NOTIFY_OK;
2213 	}
2214 #endif
2215 
2216 	if (event_dev)
2217 		parent = event_dev->dev.parent;
2218 
2219 	if (parent && parent->driver == &cxgb4_driver.driver) {
2220 		switch (event) {
2221 		case NETDEV_UP:
2222 			cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2223 			break;
2224 		case NETDEV_DOWN:
2225 			cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2226 			break;
2227 		default:
2228 			break;
2229 		}
2230 	}
2231 	return NOTIFY_OK;
2232 }
2233 
2234 static bool inet6addr_registered;
2235 static struct notifier_block cxgb4_inet6addr_notifier = {
2236 	.notifier_call = cxgb4_inet6addr_handler
2237 };
2238 
2239 static void update_clip(const struct adapter *adap)
2240 {
2241 	int i;
2242 	struct net_device *dev;
2243 	int ret;
2244 
2245 	rcu_read_lock();
2246 
2247 	for (i = 0; i < MAX_NPORTS; i++) {
2248 		dev = adap->port[i];
2249 		ret = 0;
2250 
2251 		if (dev)
2252 			ret = cxgb4_update_root_dev_clip(dev);
2253 
2254 		if (ret < 0)
2255 			break;
2256 	}
2257 	rcu_read_unlock();
2258 }
2259 #endif /* IS_ENABLED(CONFIG_IPV6) */
2260 
2261 /**
2262  *	cxgb_up - enable the adapter
2263  *	@adap: adapter being enabled
2264  *
2265  *	Called when the first port is enabled, this function performs the
2266  *	actions necessary to make an adapter operational, such as completing
2267  *	the initialization of HW modules, and enabling interrupts.
2268  *
2269  *	Must be called with the rtnl lock held.
2270  */
2271 static int cxgb_up(struct adapter *adap)
2272 {
2273 	int err;
2274 
2275 	mutex_lock(&uld_mutex);
2276 	err = setup_sge_queues(adap);
2277 	if (err)
2278 		goto rel_lock;
2279 	err = setup_rss(adap);
2280 	if (err)
2281 		goto freeq;
2282 
2283 	if (adap->flags & CXGB4_USING_MSIX) {
2284 		name_msix_vecs(adap);
2285 		err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2286 				  adap->msix_info[0].desc, adap);
2287 		if (err)
2288 			goto irq_err;
2289 		err = request_msix_queue_irqs(adap);
2290 		if (err) {
2291 			free_irq(adap->msix_info[0].vec, adap);
2292 			goto irq_err;
2293 		}
2294 	} else {
2295 		err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2296 				  (adap->flags & CXGB4_USING_MSI) ? 0
2297 								  : IRQF_SHARED,
2298 				  adap->port[0]->name, adap);
2299 		if (err)
2300 			goto irq_err;
2301 	}
2302 
2303 	enable_rx(adap);
2304 	t4_sge_start(adap);
2305 	t4_intr_enable(adap);
2306 	adap->flags |= CXGB4_FULL_INIT_DONE;
2307 	mutex_unlock(&uld_mutex);
2308 
2309 	notify_ulds(adap, CXGB4_STATE_UP);
2310 #if IS_ENABLED(CONFIG_IPV6)
2311 	update_clip(adap);
2312 #endif
2313 	return err;
2314 
2315  irq_err:
2316 	dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2317  freeq:
2318 	t4_free_sge_resources(adap);
2319  rel_lock:
2320 	mutex_unlock(&uld_mutex);
2321 	return err;
2322 }
2323 
2324 static void cxgb_down(struct adapter *adapter)
2325 {
2326 	cancel_work_sync(&adapter->tid_release_task);
2327 	cancel_work_sync(&adapter->db_full_task);
2328 	cancel_work_sync(&adapter->db_drop_task);
2329 	adapter->tid_release_task_busy = false;
2330 	adapter->tid_release_head = NULL;
2331 
2332 	t4_sge_stop(adapter);
2333 	t4_free_sge_resources(adapter);
2334 
2335 	adapter->flags &= ~CXGB4_FULL_INIT_DONE;
2336 }
2337 
2338 /*
2339  * net_device operations
2340  */
2341 static int cxgb_open(struct net_device *dev)
2342 {
2343 	int err;
2344 	struct port_info *pi = netdev_priv(dev);
2345 	struct adapter *adapter = pi->adapter;
2346 
2347 	netif_carrier_off(dev);
2348 
2349 	if (!(adapter->flags & CXGB4_FULL_INIT_DONE)) {
2350 		err = cxgb_up(adapter);
2351 		if (err < 0)
2352 			return err;
2353 	}
2354 
2355 	/* It's possible that the basic port information could have
2356 	 * changed since we first read it.
2357 	 */
2358 	err = t4_update_port_info(pi);
2359 	if (err < 0)
2360 		return err;
2361 
2362 	err = link_start(dev);
2363 	if (!err)
2364 		netif_tx_start_all_queues(dev);
2365 	return err;
2366 }
2367 
2368 static int cxgb_close(struct net_device *dev)
2369 {
2370 	struct port_info *pi = netdev_priv(dev);
2371 	struct adapter *adapter = pi->adapter;
2372 	int ret;
2373 
2374 	netif_tx_stop_all_queues(dev);
2375 	netif_carrier_off(dev);
2376 	ret = t4_enable_pi_params(adapter, adapter->pf, pi,
2377 				  false, false, false);
2378 #ifdef CONFIG_CHELSIO_T4_DCB
2379 	cxgb4_dcb_reset(dev);
2380 	dcb_tx_queue_prio_enable(dev, false);
2381 #endif
2382 	return ret;
2383 }
2384 
2385 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2386 		__be32 sip, __be16 sport, __be16 vlan,
2387 		unsigned int queue, unsigned char port, unsigned char mask)
2388 {
2389 	int ret;
2390 	struct filter_entry *f;
2391 	struct adapter *adap;
2392 	int i;
2393 	u8 *val;
2394 
2395 	adap = netdev2adap(dev);
2396 
2397 	/* Adjust stid to correct filter index */
2398 	stid -= adap->tids.sftid_base;
2399 	stid += adap->tids.nftids;
2400 
2401 	/* Check to make sure the filter requested is writable ...
2402 	 */
2403 	f = &adap->tids.ftid_tab[stid];
2404 	ret = writable_filter(f);
2405 	if (ret)
2406 		return ret;
2407 
2408 	/* Clear out any old resources being used by the filter before
2409 	 * we start constructing the new filter.
2410 	 */
2411 	if (f->valid)
2412 		clear_filter(adap, f);
2413 
2414 	/* Clear out filter specifications */
2415 	memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2416 	f->fs.val.lport = cpu_to_be16(sport);
2417 	f->fs.mask.lport  = ~0;
2418 	val = (u8 *)&sip;
2419 	if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2420 		for (i = 0; i < 4; i++) {
2421 			f->fs.val.lip[i] = val[i];
2422 			f->fs.mask.lip[i] = ~0;
2423 		}
2424 		if (adap->params.tp.vlan_pri_map & PORT_F) {
2425 			f->fs.val.iport = port;
2426 			f->fs.mask.iport = mask;
2427 		}
2428 	}
2429 
2430 	if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2431 		f->fs.val.proto = IPPROTO_TCP;
2432 		f->fs.mask.proto = ~0;
2433 	}
2434 
2435 	f->fs.dirsteer = 1;
2436 	f->fs.iq = queue;
2437 	/* Mark filter as locked */
2438 	f->locked = 1;
2439 	f->fs.rpttid = 1;
2440 
2441 	/* Save the actual tid. We need this to get the corresponding
2442 	 * filter entry structure in filter_rpl.
2443 	 */
2444 	f->tid = stid + adap->tids.ftid_base;
2445 	ret = set_filter_wr(adap, stid);
2446 	if (ret) {
2447 		clear_filter(adap, f);
2448 		return ret;
2449 	}
2450 
2451 	return 0;
2452 }
2453 EXPORT_SYMBOL(cxgb4_create_server_filter);
2454 
2455 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2456 		unsigned int queue, bool ipv6)
2457 {
2458 	struct filter_entry *f;
2459 	struct adapter *adap;
2460 
2461 	adap = netdev2adap(dev);
2462 
2463 	/* Adjust stid to correct filter index */
2464 	stid -= adap->tids.sftid_base;
2465 	stid += adap->tids.nftids;
2466 
2467 	f = &adap->tids.ftid_tab[stid];
2468 	/* Unlock the filter */
2469 	f->locked = 0;
2470 
2471 	return delete_filter(adap, stid);
2472 }
2473 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2474 
2475 static void cxgb_get_stats(struct net_device *dev,
2476 			   struct rtnl_link_stats64 *ns)
2477 {
2478 	struct port_stats stats;
2479 	struct port_info *p = netdev_priv(dev);
2480 	struct adapter *adapter = p->adapter;
2481 
2482 	/* Block retrieving statistics during EEH error
2483 	 * recovery. Otherwise, the recovery might fail
2484 	 * and the PCI device will be removed permanently
2485 	 */
2486 	spin_lock(&adapter->stats_lock);
2487 	if (!netif_device_present(dev)) {
2488 		spin_unlock(&adapter->stats_lock);
2489 		return;
2490 	}
2491 	t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2492 				 &p->stats_base);
2493 	spin_unlock(&adapter->stats_lock);
2494 
2495 	ns->tx_bytes   = stats.tx_octets;
2496 	ns->tx_packets = stats.tx_frames;
2497 	ns->rx_bytes   = stats.rx_octets;
2498 	ns->rx_packets = stats.rx_frames;
2499 	ns->multicast  = stats.rx_mcast_frames;
2500 
2501 	/* detailed rx_errors */
2502 	ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2503 			       stats.rx_runt;
2504 	ns->rx_over_errors   = 0;
2505 	ns->rx_crc_errors    = stats.rx_fcs_err;
2506 	ns->rx_frame_errors  = stats.rx_symbol_err;
2507 	ns->rx_dropped	     = stats.rx_ovflow0 + stats.rx_ovflow1 +
2508 			       stats.rx_ovflow2 + stats.rx_ovflow3 +
2509 			       stats.rx_trunc0 + stats.rx_trunc1 +
2510 			       stats.rx_trunc2 + stats.rx_trunc3;
2511 	ns->rx_missed_errors = 0;
2512 
2513 	/* detailed tx_errors */
2514 	ns->tx_aborted_errors   = 0;
2515 	ns->tx_carrier_errors   = 0;
2516 	ns->tx_fifo_errors      = 0;
2517 	ns->tx_heartbeat_errors = 0;
2518 	ns->tx_window_errors    = 0;
2519 
2520 	ns->tx_errors = stats.tx_error_frames;
2521 	ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2522 		ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2523 }
2524 
2525 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2526 {
2527 	unsigned int mbox;
2528 	int ret = 0, prtad, devad;
2529 	struct port_info *pi = netdev_priv(dev);
2530 	struct adapter *adapter = pi->adapter;
2531 	struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2532 
2533 	switch (cmd) {
2534 	case SIOCGMIIPHY:
2535 		if (pi->mdio_addr < 0)
2536 			return -EOPNOTSUPP;
2537 		data->phy_id = pi->mdio_addr;
2538 		break;
2539 	case SIOCGMIIREG:
2540 	case SIOCSMIIREG:
2541 		if (mdio_phy_id_is_c45(data->phy_id)) {
2542 			prtad = mdio_phy_id_prtad(data->phy_id);
2543 			devad = mdio_phy_id_devad(data->phy_id);
2544 		} else if (data->phy_id < 32) {
2545 			prtad = data->phy_id;
2546 			devad = 0;
2547 			data->reg_num &= 0x1f;
2548 		} else
2549 			return -EINVAL;
2550 
2551 		mbox = pi->adapter->pf;
2552 		if (cmd == SIOCGMIIREG)
2553 			ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2554 					 data->reg_num, &data->val_out);
2555 		else
2556 			ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2557 					 data->reg_num, data->val_in);
2558 		break;
2559 	case SIOCGHWTSTAMP:
2560 		return copy_to_user(req->ifr_data, &pi->tstamp_config,
2561 				    sizeof(pi->tstamp_config)) ?
2562 			-EFAULT : 0;
2563 	case SIOCSHWTSTAMP:
2564 		if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2565 				   sizeof(pi->tstamp_config)))
2566 			return -EFAULT;
2567 
2568 		if (!is_t4(adapter->params.chip)) {
2569 			switch (pi->tstamp_config.tx_type) {
2570 			case HWTSTAMP_TX_OFF:
2571 			case HWTSTAMP_TX_ON:
2572 				break;
2573 			default:
2574 				return -ERANGE;
2575 			}
2576 
2577 			switch (pi->tstamp_config.rx_filter) {
2578 			case HWTSTAMP_FILTER_NONE:
2579 				pi->rxtstamp = false;
2580 				break;
2581 			case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2582 			case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2583 				cxgb4_ptprx_timestamping(pi, pi->port_id,
2584 							 PTP_TS_L4);
2585 				break;
2586 			case HWTSTAMP_FILTER_PTP_V2_EVENT:
2587 				cxgb4_ptprx_timestamping(pi, pi->port_id,
2588 							 PTP_TS_L2_L4);
2589 				break;
2590 			case HWTSTAMP_FILTER_ALL:
2591 			case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2592 			case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2593 			case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2594 			case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2595 				pi->rxtstamp = true;
2596 				break;
2597 			default:
2598 				pi->tstamp_config.rx_filter =
2599 					HWTSTAMP_FILTER_NONE;
2600 				return -ERANGE;
2601 			}
2602 
2603 			if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
2604 			    (pi->tstamp_config.rx_filter ==
2605 				HWTSTAMP_FILTER_NONE)) {
2606 				if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
2607 					pi->ptp_enable = false;
2608 			}
2609 
2610 			if (pi->tstamp_config.rx_filter !=
2611 				HWTSTAMP_FILTER_NONE) {
2612 				if (cxgb4_ptp_redirect_rx_packet(adapter,
2613 								 pi) >= 0)
2614 					pi->ptp_enable = true;
2615 			}
2616 		} else {
2617 			/* For T4 Adapters */
2618 			switch (pi->tstamp_config.rx_filter) {
2619 			case HWTSTAMP_FILTER_NONE:
2620 			pi->rxtstamp = false;
2621 			break;
2622 			case HWTSTAMP_FILTER_ALL:
2623 			pi->rxtstamp = true;
2624 			break;
2625 			default:
2626 			pi->tstamp_config.rx_filter =
2627 			HWTSTAMP_FILTER_NONE;
2628 			return -ERANGE;
2629 			}
2630 		}
2631 		return copy_to_user(req->ifr_data, &pi->tstamp_config,
2632 				    sizeof(pi->tstamp_config)) ?
2633 			-EFAULT : 0;
2634 	default:
2635 		return -EOPNOTSUPP;
2636 	}
2637 	return ret;
2638 }
2639 
2640 static void cxgb_set_rxmode(struct net_device *dev)
2641 {
2642 	/* unfortunately we can't return errors to the stack */
2643 	set_rxmode(dev, -1, false);
2644 }
2645 
2646 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2647 {
2648 	int ret;
2649 	struct port_info *pi = netdev_priv(dev);
2650 
2651 	ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2652 			    -1, -1, -1, true);
2653 	if (!ret)
2654 		dev->mtu = new_mtu;
2655 	return ret;
2656 }
2657 
2658 #ifdef CONFIG_PCI_IOV
2659 static int cxgb4_mgmt_open(struct net_device *dev)
2660 {
2661 	/* Turn carrier off since we don't have to transmit anything on this
2662 	 * interface.
2663 	 */
2664 	netif_carrier_off(dev);
2665 	return 0;
2666 }
2667 
2668 /* Fill MAC address that will be assigned by the FW */
2669 static void cxgb4_mgmt_fill_vf_station_mac_addr(struct adapter *adap)
2670 {
2671 	u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2672 	unsigned int i, vf, nvfs;
2673 	u16 a, b;
2674 	int err;
2675 	u8 *na;
2676 
2677 	adap->params.pci.vpd_cap_addr = pci_find_capability(adap->pdev,
2678 							    PCI_CAP_ID_VPD);
2679 	err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2680 	if (err)
2681 		return;
2682 
2683 	na = adap->params.vpd.na;
2684 	for (i = 0; i < ETH_ALEN; i++)
2685 		hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2686 			      hex2val(na[2 * i + 1]));
2687 
2688 	a = (hw_addr[0] << 8) | hw_addr[1];
2689 	b = (hw_addr[1] << 8) | hw_addr[2];
2690 	a ^= b;
2691 	a |= 0x0200;    /* locally assigned Ethernet MAC address */
2692 	a &= ~0x0100;   /* not a multicast Ethernet MAC address */
2693 	macaddr[0] = a >> 8;
2694 	macaddr[1] = a & 0xff;
2695 
2696 	for (i = 2; i < 5; i++)
2697 		macaddr[i] = hw_addr[i + 1];
2698 
2699 	for (vf = 0, nvfs = pci_sriov_get_totalvfs(adap->pdev);
2700 		vf < nvfs; vf++) {
2701 		macaddr[5] = adap->pf * nvfs + vf;
2702 		ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, macaddr);
2703 	}
2704 }
2705 
2706 static int cxgb4_mgmt_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2707 {
2708 	struct port_info *pi = netdev_priv(dev);
2709 	struct adapter *adap = pi->adapter;
2710 	int ret;
2711 
2712 	/* verify MAC addr is valid */
2713 	if (!is_valid_ether_addr(mac)) {
2714 		dev_err(pi->adapter->pdev_dev,
2715 			"Invalid Ethernet address %pM for VF %d\n",
2716 			mac, vf);
2717 		return -EINVAL;
2718 	}
2719 
2720 	dev_info(pi->adapter->pdev_dev,
2721 		 "Setting MAC %pM on VF %d\n", mac, vf);
2722 	ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2723 	if (!ret)
2724 		ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2725 	return ret;
2726 }
2727 
2728 static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
2729 				    int vf, struct ifla_vf_info *ivi)
2730 {
2731 	struct port_info *pi = netdev_priv(dev);
2732 	struct adapter *adap = pi->adapter;
2733 	struct vf_info *vfinfo;
2734 
2735 	if (vf >= adap->num_vfs)
2736 		return -EINVAL;
2737 	vfinfo = &adap->vfinfo[vf];
2738 
2739 	ivi->vf = vf;
2740 	ivi->max_tx_rate = vfinfo->tx_rate;
2741 	ivi->min_tx_rate = 0;
2742 	ether_addr_copy(ivi->mac, vfinfo->vf_mac_addr);
2743 	ivi->vlan = vfinfo->vlan;
2744 	ivi->linkstate = vfinfo->link_state;
2745 	return 0;
2746 }
2747 
2748 static int cxgb4_mgmt_get_phys_port_id(struct net_device *dev,
2749 				       struct netdev_phys_item_id *ppid)
2750 {
2751 	struct port_info *pi = netdev_priv(dev);
2752 	unsigned int phy_port_id;
2753 
2754 	phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2755 	ppid->id_len = sizeof(phy_port_id);
2756 	memcpy(ppid->id, &phy_port_id, ppid->id_len);
2757 	return 0;
2758 }
2759 
2760 static int cxgb4_mgmt_set_vf_rate(struct net_device *dev, int vf,
2761 				  int min_tx_rate, int max_tx_rate)
2762 {
2763 	struct port_info *pi = netdev_priv(dev);
2764 	struct adapter *adap = pi->adapter;
2765 	unsigned int link_ok, speed, mtu;
2766 	u32 fw_pfvf, fw_class;
2767 	int class_id = vf;
2768 	int ret;
2769 	u16 pktsize;
2770 
2771 	if (vf >= adap->num_vfs)
2772 		return -EINVAL;
2773 
2774 	if (min_tx_rate) {
2775 		dev_err(adap->pdev_dev,
2776 			"Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2777 			min_tx_rate, vf);
2778 		return -EINVAL;
2779 	}
2780 
2781 	if (max_tx_rate == 0) {
2782 		/* unbind VF to to any Traffic Class */
2783 		fw_pfvf =
2784 		    (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2785 		     FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2786 		fw_class = 0xffffffff;
2787 		ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
2788 				    &fw_pfvf, &fw_class);
2789 		if (ret) {
2790 			dev_err(adap->pdev_dev,
2791 				"Err %d in unbinding PF %d VF %d from TX Rate Limiting\n",
2792 				ret, adap->pf, vf);
2793 			return -EINVAL;
2794 		}
2795 		dev_info(adap->pdev_dev,
2796 			 "PF %d VF %d is unbound from TX Rate Limiting\n",
2797 			 adap->pf, vf);
2798 		adap->vfinfo[vf].tx_rate = 0;
2799 		return 0;
2800 	}
2801 
2802 	ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
2803 	if (ret != FW_SUCCESS) {
2804 		dev_err(adap->pdev_dev,
2805 			"Failed to get link information for VF %d\n", vf);
2806 		return -EINVAL;
2807 	}
2808 
2809 	if (!link_ok) {
2810 		dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
2811 		return -EINVAL;
2812 	}
2813 
2814 	if (max_tx_rate > speed) {
2815 		dev_err(adap->pdev_dev,
2816 			"Max tx rate %d for VF %d can't be > link-speed %u",
2817 			max_tx_rate, vf, speed);
2818 		return -EINVAL;
2819 	}
2820 
2821 	pktsize = mtu;
2822 	/* subtract ethhdr size and 4 bytes crc since, f/w appends it */
2823 	pktsize = pktsize - sizeof(struct ethhdr) - 4;
2824 	/* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
2825 	pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
2826 	/* configure Traffic Class for rate-limiting */
2827 	ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
2828 			      SCHED_CLASS_LEVEL_CL_RL,
2829 			      SCHED_CLASS_MODE_CLASS,
2830 			      SCHED_CLASS_RATEUNIT_BITS,
2831 			      SCHED_CLASS_RATEMODE_ABS,
2832 			      pi->tx_chan, class_id, 0,
2833 			      max_tx_rate * 1000, 0, pktsize);
2834 	if (ret) {
2835 		dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
2836 			ret);
2837 		return -EINVAL;
2838 	}
2839 	dev_info(adap->pdev_dev,
2840 		 "Class %d with MSS %u configured with rate %u\n",
2841 		 class_id, pktsize, max_tx_rate);
2842 
2843 	/* bind VF to configured Traffic Class */
2844 	fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2845 		   FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2846 	fw_class = class_id;
2847 	ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
2848 			    &fw_class);
2849 	if (ret) {
2850 		dev_err(adap->pdev_dev,
2851 			"Err %d in binding PF %d VF %d to Traffic Class %d\n",
2852 			ret, adap->pf, vf, class_id);
2853 		return -EINVAL;
2854 	}
2855 	dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
2856 		 adap->pf, vf, class_id);
2857 	adap->vfinfo[vf].tx_rate = max_tx_rate;
2858 	return 0;
2859 }
2860 
2861 static int cxgb4_mgmt_set_vf_vlan(struct net_device *dev, int vf,
2862 				  u16 vlan, u8 qos, __be16 vlan_proto)
2863 {
2864 	struct port_info *pi = netdev_priv(dev);
2865 	struct adapter *adap = pi->adapter;
2866 	int ret;
2867 
2868 	if (vf >= adap->num_vfs || vlan > 4095 || qos > 7)
2869 		return -EINVAL;
2870 
2871 	if (vlan_proto != htons(ETH_P_8021Q) || qos != 0)
2872 		return -EPROTONOSUPPORT;
2873 
2874 	ret = t4_set_vlan_acl(adap, adap->mbox, vf + 1, vlan);
2875 	if (!ret) {
2876 		adap->vfinfo[vf].vlan = vlan;
2877 		return 0;
2878 	}
2879 
2880 	dev_err(adap->pdev_dev, "Err %d %s VLAN ACL for PF/VF %d/%d\n",
2881 		ret, (vlan ? "setting" : "clearing"), adap->pf, vf);
2882 	return ret;
2883 }
2884 
2885 static int cxgb4_mgmt_set_vf_link_state(struct net_device *dev, int vf,
2886 					int link)
2887 {
2888 	struct port_info *pi = netdev_priv(dev);
2889 	struct adapter *adap = pi->adapter;
2890 	u32 param, val;
2891 	int ret = 0;
2892 
2893 	if (vf >= adap->num_vfs)
2894 		return -EINVAL;
2895 
2896 	switch (link) {
2897 	case IFLA_VF_LINK_STATE_AUTO:
2898 		val = FW_VF_LINK_STATE_AUTO;
2899 		break;
2900 
2901 	case IFLA_VF_LINK_STATE_ENABLE:
2902 		val = FW_VF_LINK_STATE_ENABLE;
2903 		break;
2904 
2905 	case IFLA_VF_LINK_STATE_DISABLE:
2906 		val = FW_VF_LINK_STATE_DISABLE;
2907 		break;
2908 
2909 	default:
2910 		return -EINVAL;
2911 	}
2912 
2913 	param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2914 		 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_LINK_STATE));
2915 	ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
2916 			    &param, &val);
2917 	if (ret) {
2918 		dev_err(adap->pdev_dev,
2919 			"Error %d in setting PF %d VF %d link state\n",
2920 			ret, adap->pf, vf);
2921 		return -EINVAL;
2922 	}
2923 
2924 	adap->vfinfo[vf].link_state = link;
2925 	return ret;
2926 }
2927 #endif /* CONFIG_PCI_IOV */
2928 
2929 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2930 {
2931 	int ret;
2932 	struct sockaddr *addr = p;
2933 	struct port_info *pi = netdev_priv(dev);
2934 
2935 	if (!is_valid_ether_addr(addr->sa_data))
2936 		return -EADDRNOTAVAIL;
2937 
2938 	ret = cxgb4_change_mac(pi, pi->viid, &pi->xact_addr_filt,
2939 			       addr->sa_data, true, &pi->smt_idx);
2940 	if (ret < 0)
2941 		return ret;
2942 
2943 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2944 	pi->xact_addr_filt = ret;
2945 	return 0;
2946 }
2947 
2948 #ifdef CONFIG_NET_POLL_CONTROLLER
2949 static void cxgb_netpoll(struct net_device *dev)
2950 {
2951 	struct port_info *pi = netdev_priv(dev);
2952 	struct adapter *adap = pi->adapter;
2953 
2954 	if (adap->flags & CXGB4_USING_MSIX) {
2955 		int i;
2956 		struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2957 
2958 		for (i = pi->nqsets; i; i--, rx++)
2959 			t4_sge_intr_msix(0, &rx->rspq);
2960 	} else
2961 		t4_intr_handler(adap)(0, adap);
2962 }
2963 #endif
2964 
2965 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2966 {
2967 	struct port_info *pi = netdev_priv(dev);
2968 	struct adapter *adap = pi->adapter;
2969 	struct sched_class *e;
2970 	struct ch_sched_params p;
2971 	struct ch_sched_queue qe;
2972 	u32 req_rate;
2973 	int err = 0;
2974 
2975 	if (!can_sched(dev))
2976 		return -ENOTSUPP;
2977 
2978 	if (index < 0 || index > pi->nqsets - 1)
2979 		return -EINVAL;
2980 
2981 	if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
2982 		dev_err(adap->pdev_dev,
2983 			"Failed to rate limit on queue %d. Link Down?\n",
2984 			index);
2985 		return -EINVAL;
2986 	}
2987 
2988 	/* Convert from Mbps to Kbps */
2989 	req_rate = rate * 1000;
2990 
2991 	/* Max rate is 100 Gbps */
2992 	if (req_rate > SCHED_MAX_RATE_KBPS) {
2993 		dev_err(adap->pdev_dev,
2994 			"Invalid rate %u Mbps, Max rate is %u Mbps\n",
2995 			rate, SCHED_MAX_RATE_KBPS / 1000);
2996 		return -ERANGE;
2997 	}
2998 
2999 	/* First unbind the queue from any existing class */
3000 	memset(&qe, 0, sizeof(qe));
3001 	qe.queue = index;
3002 	qe.class = SCHED_CLS_NONE;
3003 
3004 	err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
3005 	if (err) {
3006 		dev_err(adap->pdev_dev,
3007 			"Unbinding Queue %d on port %d fail. Err: %d\n",
3008 			index, pi->port_id, err);
3009 		return err;
3010 	}
3011 
3012 	/* Queue already unbound */
3013 	if (!req_rate)
3014 		return 0;
3015 
3016 	/* Fetch any available unused or matching scheduling class */
3017 	memset(&p, 0, sizeof(p));
3018 	p.type = SCHED_CLASS_TYPE_PACKET;
3019 	p.u.params.level    = SCHED_CLASS_LEVEL_CL_RL;
3020 	p.u.params.mode     = SCHED_CLASS_MODE_CLASS;
3021 	p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
3022 	p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
3023 	p.u.params.channel  = pi->tx_chan;
3024 	p.u.params.class    = SCHED_CLS_NONE;
3025 	p.u.params.minrate  = 0;
3026 	p.u.params.maxrate  = req_rate;
3027 	p.u.params.weight   = 0;
3028 	p.u.params.pktsize  = dev->mtu;
3029 
3030 	e = cxgb4_sched_class_alloc(dev, &p);
3031 	if (!e)
3032 		return -ENOMEM;
3033 
3034 	/* Bind the queue to a scheduling class */
3035 	memset(&qe, 0, sizeof(qe));
3036 	qe.queue = index;
3037 	qe.class = e->idx;
3038 
3039 	err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
3040 	if (err)
3041 		dev_err(adap->pdev_dev,
3042 			"Queue rate limiting failed. Err: %d\n", err);
3043 	return err;
3044 }
3045 
3046 static int cxgb_setup_tc_flower(struct net_device *dev,
3047 				struct tc_cls_flower_offload *cls_flower)
3048 {
3049 	switch (cls_flower->command) {
3050 	case TC_CLSFLOWER_REPLACE:
3051 		return cxgb4_tc_flower_replace(dev, cls_flower);
3052 	case TC_CLSFLOWER_DESTROY:
3053 		return cxgb4_tc_flower_destroy(dev, cls_flower);
3054 	case TC_CLSFLOWER_STATS:
3055 		return cxgb4_tc_flower_stats(dev, cls_flower);
3056 	default:
3057 		return -EOPNOTSUPP;
3058 	}
3059 }
3060 
3061 static int cxgb_setup_tc_cls_u32(struct net_device *dev,
3062 				 struct tc_cls_u32_offload *cls_u32)
3063 {
3064 	switch (cls_u32->command) {
3065 	case TC_CLSU32_NEW_KNODE:
3066 	case TC_CLSU32_REPLACE_KNODE:
3067 		return cxgb4_config_knode(dev, cls_u32);
3068 	case TC_CLSU32_DELETE_KNODE:
3069 		return cxgb4_delete_knode(dev, cls_u32);
3070 	default:
3071 		return -EOPNOTSUPP;
3072 	}
3073 }
3074 
3075 static int cxgb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
3076 				  void *cb_priv)
3077 {
3078 	struct net_device *dev = cb_priv;
3079 	struct port_info *pi = netdev2pinfo(dev);
3080 	struct adapter *adap = netdev2adap(dev);
3081 
3082 	if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3083 		dev_err(adap->pdev_dev,
3084 			"Failed to setup tc on port %d. Link Down?\n",
3085 			pi->port_id);
3086 		return -EINVAL;
3087 	}
3088 
3089 	if (!tc_cls_can_offload_and_chain0(dev, type_data))
3090 		return -EOPNOTSUPP;
3091 
3092 	switch (type) {
3093 	case TC_SETUP_CLSU32:
3094 		return cxgb_setup_tc_cls_u32(dev, type_data);
3095 	case TC_SETUP_CLSFLOWER:
3096 		return cxgb_setup_tc_flower(dev, type_data);
3097 	default:
3098 		return -EOPNOTSUPP;
3099 	}
3100 }
3101 
3102 static int cxgb_setup_tc_block(struct net_device *dev,
3103 			       struct tc_block_offload *f)
3104 {
3105 	struct port_info *pi = netdev2pinfo(dev);
3106 
3107 	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
3108 		return -EOPNOTSUPP;
3109 
3110 	switch (f->command) {
3111 	case TC_BLOCK_BIND:
3112 		return tcf_block_cb_register(f->block, cxgb_setup_tc_block_cb,
3113 					     pi, dev, f->extack);
3114 	case TC_BLOCK_UNBIND:
3115 		tcf_block_cb_unregister(f->block, cxgb_setup_tc_block_cb, pi);
3116 		return 0;
3117 	default:
3118 		return -EOPNOTSUPP;
3119 	}
3120 }
3121 
3122 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
3123 			 void *type_data)
3124 {
3125 	switch (type) {
3126 	case TC_SETUP_BLOCK:
3127 		return cxgb_setup_tc_block(dev, type_data);
3128 	default:
3129 		return -EOPNOTSUPP;
3130 	}
3131 }
3132 
3133 static void cxgb_del_udp_tunnel(struct net_device *netdev,
3134 				struct udp_tunnel_info *ti)
3135 {
3136 	struct port_info *pi = netdev_priv(netdev);
3137 	struct adapter *adapter = pi->adapter;
3138 	unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3139 	u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3140 	int ret = 0, i;
3141 
3142 	if (chip_ver < CHELSIO_T6)
3143 		return;
3144 
3145 	switch (ti->type) {
3146 	case UDP_TUNNEL_TYPE_VXLAN:
3147 		if (!adapter->vxlan_port_cnt ||
3148 		    adapter->vxlan_port != ti->port)
3149 			return; /* Invalid VxLAN destination port */
3150 
3151 		adapter->vxlan_port_cnt--;
3152 		if (adapter->vxlan_port_cnt)
3153 			return;
3154 
3155 		adapter->vxlan_port = 0;
3156 		t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
3157 		break;
3158 	case UDP_TUNNEL_TYPE_GENEVE:
3159 		if (!adapter->geneve_port_cnt ||
3160 		    adapter->geneve_port != ti->port)
3161 			return; /* Invalid GENEVE destination port */
3162 
3163 		adapter->geneve_port_cnt--;
3164 		if (adapter->geneve_port_cnt)
3165 			return;
3166 
3167 		adapter->geneve_port = 0;
3168 		t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A, 0);
3169 		break;
3170 	default:
3171 		return;
3172 	}
3173 
3174 	/* Matchall mac entries can be deleted only after all tunnel ports
3175 	 * are brought down or removed.
3176 	 */
3177 	if (!adapter->rawf_cnt)
3178 		return;
3179 	for_each_port(adapter, i) {
3180 		pi = adap2pinfo(adapter, i);
3181 		ret = t4_free_raw_mac_filt(adapter, pi->viid,
3182 					   match_all_mac, match_all_mac,
3183 					   adapter->rawf_start +
3184 					    pi->port_id,
3185 					   1, pi->port_id, false);
3186 		if (ret < 0) {
3187 			netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
3188 				    i);
3189 			return;
3190 		}
3191 		atomic_dec(&adapter->mps_encap[adapter->rawf_start +
3192 			   pi->port_id].refcnt);
3193 	}
3194 }
3195 
3196 static void cxgb_add_udp_tunnel(struct net_device *netdev,
3197 				struct udp_tunnel_info *ti)
3198 {
3199 	struct port_info *pi = netdev_priv(netdev);
3200 	struct adapter *adapter = pi->adapter;
3201 	unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3202 	u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3203 	int i, ret;
3204 
3205 	if (chip_ver < CHELSIO_T6 || !adapter->rawf_cnt)
3206 		return;
3207 
3208 	switch (ti->type) {
3209 	case UDP_TUNNEL_TYPE_VXLAN:
3210 		/* Callback for adding vxlan port can be called with the same
3211 		 * port for both IPv4 and IPv6. We should not disable the
3212 		 * offloading when the same port for both protocols is added
3213 		 * and later one of them is removed.
3214 		 */
3215 		if (adapter->vxlan_port_cnt &&
3216 		    adapter->vxlan_port == ti->port) {
3217 			adapter->vxlan_port_cnt++;
3218 			return;
3219 		}
3220 
3221 		/* We will support only one VxLAN port */
3222 		if (adapter->vxlan_port_cnt) {
3223 			netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3224 				    be16_to_cpu(adapter->vxlan_port),
3225 				    be16_to_cpu(ti->port));
3226 			return;
3227 		}
3228 
3229 		adapter->vxlan_port = ti->port;
3230 		adapter->vxlan_port_cnt = 1;
3231 
3232 		t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
3233 			     VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
3234 		break;
3235 	case UDP_TUNNEL_TYPE_GENEVE:
3236 		if (adapter->geneve_port_cnt &&
3237 		    adapter->geneve_port == ti->port) {
3238 			adapter->geneve_port_cnt++;
3239 			return;
3240 		}
3241 
3242 		/* We will support only one GENEVE port */
3243 		if (adapter->geneve_port_cnt) {
3244 			netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3245 				    be16_to_cpu(adapter->geneve_port),
3246 				    be16_to_cpu(ti->port));
3247 			return;
3248 		}
3249 
3250 		adapter->geneve_port = ti->port;
3251 		adapter->geneve_port_cnt = 1;
3252 
3253 		t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A,
3254 			     GENEVE_V(be16_to_cpu(ti->port)) | GENEVE_EN_F);
3255 		break;
3256 	default:
3257 		return;
3258 	}
3259 
3260 	/* Create a 'match all' mac filter entry for inner mac,
3261 	 * if raw mac interface is supported. Once the linux kernel provides
3262 	 * driver entry points for adding/deleting the inner mac addresses,
3263 	 * we will remove this 'match all' entry and fallback to adding
3264 	 * exact match filters.
3265 	 */
3266 	for_each_port(adapter, i) {
3267 		pi = adap2pinfo(adapter, i);
3268 
3269 		ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
3270 					    match_all_mac,
3271 					    match_all_mac,
3272 					    adapter->rawf_start +
3273 					    pi->port_id,
3274 					    1, pi->port_id, false);
3275 		if (ret < 0) {
3276 			netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
3277 				    be16_to_cpu(ti->port));
3278 			cxgb_del_udp_tunnel(netdev, ti);
3279 			return;
3280 		}
3281 		atomic_inc(&adapter->mps_encap[ret].refcnt);
3282 	}
3283 }
3284 
3285 static netdev_features_t cxgb_features_check(struct sk_buff *skb,
3286 					     struct net_device *dev,
3287 					     netdev_features_t features)
3288 {
3289 	struct port_info *pi = netdev_priv(dev);
3290 	struct adapter *adapter = pi->adapter;
3291 
3292 	if (CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3293 		return features;
3294 
3295 	/* Check if hw supports offload for this packet */
3296 	if (!skb->encapsulation || cxgb_encap_offload_supported(skb))
3297 		return features;
3298 
3299 	/* Offload is not supported for this encapsulated packet */
3300 	return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3301 }
3302 
3303 static netdev_features_t cxgb_fix_features(struct net_device *dev,
3304 					   netdev_features_t features)
3305 {
3306 	/* Disable GRO, if RX_CSUM is disabled */
3307 	if (!(features & NETIF_F_RXCSUM))
3308 		features &= ~NETIF_F_GRO;
3309 
3310 	return features;
3311 }
3312 
3313 static const struct net_device_ops cxgb4_netdev_ops = {
3314 	.ndo_open             = cxgb_open,
3315 	.ndo_stop             = cxgb_close,
3316 	.ndo_start_xmit       = t4_start_xmit,
3317 	.ndo_select_queue     =	cxgb_select_queue,
3318 	.ndo_get_stats64      = cxgb_get_stats,
3319 	.ndo_set_rx_mode      = cxgb_set_rxmode,
3320 	.ndo_set_mac_address  = cxgb_set_mac_addr,
3321 	.ndo_set_features     = cxgb_set_features,
3322 	.ndo_validate_addr    = eth_validate_addr,
3323 	.ndo_do_ioctl         = cxgb_ioctl,
3324 	.ndo_change_mtu       = cxgb_change_mtu,
3325 #ifdef CONFIG_NET_POLL_CONTROLLER
3326 	.ndo_poll_controller  = cxgb_netpoll,
3327 #endif
3328 #ifdef CONFIG_CHELSIO_T4_FCOE
3329 	.ndo_fcoe_enable      = cxgb_fcoe_enable,
3330 	.ndo_fcoe_disable     = cxgb_fcoe_disable,
3331 #endif /* CONFIG_CHELSIO_T4_FCOE */
3332 	.ndo_set_tx_maxrate   = cxgb_set_tx_maxrate,
3333 	.ndo_setup_tc         = cxgb_setup_tc,
3334 	.ndo_udp_tunnel_add   = cxgb_add_udp_tunnel,
3335 	.ndo_udp_tunnel_del   = cxgb_del_udp_tunnel,
3336 	.ndo_features_check   = cxgb_features_check,
3337 	.ndo_fix_features     = cxgb_fix_features,
3338 };
3339 
3340 #ifdef CONFIG_PCI_IOV
3341 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
3342 	.ndo_open               = cxgb4_mgmt_open,
3343 	.ndo_set_vf_mac         = cxgb4_mgmt_set_vf_mac,
3344 	.ndo_get_vf_config      = cxgb4_mgmt_get_vf_config,
3345 	.ndo_set_vf_rate        = cxgb4_mgmt_set_vf_rate,
3346 	.ndo_get_phys_port_id   = cxgb4_mgmt_get_phys_port_id,
3347 	.ndo_set_vf_vlan        = cxgb4_mgmt_set_vf_vlan,
3348 	.ndo_set_vf_link_state	= cxgb4_mgmt_set_vf_link_state,
3349 };
3350 #endif
3351 
3352 static void cxgb4_mgmt_get_drvinfo(struct net_device *dev,
3353 				   struct ethtool_drvinfo *info)
3354 {
3355 	struct adapter *adapter = netdev2adap(dev);
3356 
3357 	strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
3358 	strlcpy(info->version, cxgb4_driver_version,
3359 		sizeof(info->version));
3360 	strlcpy(info->bus_info, pci_name(adapter->pdev),
3361 		sizeof(info->bus_info));
3362 }
3363 
3364 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
3365 	.get_drvinfo       = cxgb4_mgmt_get_drvinfo,
3366 };
3367 
3368 static void notify_fatal_err(struct work_struct *work)
3369 {
3370 	struct adapter *adap;
3371 
3372 	adap = container_of(work, struct adapter, fatal_err_notify_task);
3373 	notify_ulds(adap, CXGB4_STATE_FATAL_ERROR);
3374 }
3375 
3376 void t4_fatal_err(struct adapter *adap)
3377 {
3378 	int port;
3379 
3380 	if (pci_channel_offline(adap->pdev))
3381 		return;
3382 
3383 	/* Disable the SGE since ULDs are going to free resources that
3384 	 * could be exposed to the adapter.  RDMA MWs for example...
3385 	 */
3386 	t4_shutdown_adapter(adap);
3387 	for_each_port(adap, port) {
3388 		struct net_device *dev = adap->port[port];
3389 
3390 		/* If we get here in very early initialization the network
3391 		 * devices may not have been set up yet.
3392 		 */
3393 		if (!dev)
3394 			continue;
3395 
3396 		netif_tx_stop_all_queues(dev);
3397 		netif_carrier_off(dev);
3398 	}
3399 	dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3400 	queue_work(adap->workq, &adap->fatal_err_notify_task);
3401 }
3402 
3403 static void setup_memwin(struct adapter *adap)
3404 {
3405 	u32 nic_win_base = t4_get_util_window(adap);
3406 
3407 	t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3408 }
3409 
3410 static void setup_memwin_rdma(struct adapter *adap)
3411 {
3412 	if (adap->vres.ocq.size) {
3413 		u32 start;
3414 		unsigned int sz_kb;
3415 
3416 		start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3417 		start &= PCI_BASE_ADDRESS_MEM_MASK;
3418 		start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3419 		sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3420 		t4_write_reg(adap,
3421 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3422 			     start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3423 		t4_write_reg(adap,
3424 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3425 			     adap->vres.ocq.start);
3426 		t4_read_reg(adap,
3427 			    PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3428 	}
3429 }
3430 
3431 /* HMA Definitions */
3432 
3433 /* The maximum number of address that can be send in a single FW cmd */
3434 #define HMA_MAX_ADDR_IN_CMD	5
3435 
3436 #define HMA_PAGE_SIZE		PAGE_SIZE
3437 
3438 #define HMA_MAX_NO_FW_ADDRESS	(16 << 10)  /* FW supports 16K addresses */
3439 
3440 #define HMA_PAGE_ORDER					\
3441 	((HMA_PAGE_SIZE < HMA_MAX_NO_FW_ADDRESS) ?	\
3442 	ilog2(HMA_MAX_NO_FW_ADDRESS / HMA_PAGE_SIZE) : 0)
3443 
3444 /* The minimum and maximum possible HMA sizes that can be specified in the FW
3445  * configuration(in units of MB).
3446  */
3447 #define HMA_MIN_TOTAL_SIZE	1
3448 #define HMA_MAX_TOTAL_SIZE				\
3449 	(((HMA_PAGE_SIZE << HMA_PAGE_ORDER) *		\
3450 	  HMA_MAX_NO_FW_ADDRESS) >> 20)
3451 
3452 static void adap_free_hma_mem(struct adapter *adapter)
3453 {
3454 	struct scatterlist *iter;
3455 	struct page *page;
3456 	int i;
3457 
3458 	if (!adapter->hma.sgt)
3459 		return;
3460 
3461 	if (adapter->hma.flags & HMA_DMA_MAPPED_FLAG) {
3462 		dma_unmap_sg(adapter->pdev_dev, adapter->hma.sgt->sgl,
3463 			     adapter->hma.sgt->nents, PCI_DMA_BIDIRECTIONAL);
3464 		adapter->hma.flags &= ~HMA_DMA_MAPPED_FLAG;
3465 	}
3466 
3467 	for_each_sg(adapter->hma.sgt->sgl, iter,
3468 		    adapter->hma.sgt->orig_nents, i) {
3469 		page = sg_page(iter);
3470 		if (page)
3471 			__free_pages(page, HMA_PAGE_ORDER);
3472 	}
3473 
3474 	kfree(adapter->hma.phy_addr);
3475 	sg_free_table(adapter->hma.sgt);
3476 	kfree(adapter->hma.sgt);
3477 	adapter->hma.sgt = NULL;
3478 }
3479 
3480 static int adap_config_hma(struct adapter *adapter)
3481 {
3482 	struct scatterlist *sgl, *iter;
3483 	struct sg_table *sgt;
3484 	struct page *newpage;
3485 	unsigned int i, j, k;
3486 	u32 param, hma_size;
3487 	unsigned int ncmds;
3488 	size_t page_size;
3489 	u32 page_order;
3490 	int node, ret;
3491 
3492 	/* HMA is supported only for T6+ cards.
3493 	 * Avoid initializing HMA in kdump kernels.
3494 	 */
3495 	if (is_kdump_kernel() ||
3496 	    CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3497 		return 0;
3498 
3499 	/* Get the HMA region size required by fw */
3500 	param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3501 		 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HMA_SIZE));
3502 	ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
3503 			      1, &param, &hma_size);
3504 	/* An error means card has its own memory or HMA is not supported by
3505 	 * the firmware. Return without any errors.
3506 	 */
3507 	if (ret || !hma_size)
3508 		return 0;
3509 
3510 	if (hma_size < HMA_MIN_TOTAL_SIZE ||
3511 	    hma_size > HMA_MAX_TOTAL_SIZE) {
3512 		dev_err(adapter->pdev_dev,
3513 			"HMA size %uMB beyond bounds(%u-%lu)MB\n",
3514 			hma_size, HMA_MIN_TOTAL_SIZE, HMA_MAX_TOTAL_SIZE);
3515 		return -EINVAL;
3516 	}
3517 
3518 	page_size = HMA_PAGE_SIZE;
3519 	page_order = HMA_PAGE_ORDER;
3520 	adapter->hma.sgt = kzalloc(sizeof(*adapter->hma.sgt), GFP_KERNEL);
3521 	if (unlikely(!adapter->hma.sgt)) {
3522 		dev_err(adapter->pdev_dev, "HMA SG table allocation failed\n");
3523 		return -ENOMEM;
3524 	}
3525 	sgt = adapter->hma.sgt;
3526 	/* FW returned value will be in MB's
3527 	 */
3528 	sgt->orig_nents = (hma_size << 20) / (page_size << page_order);
3529 	if (sg_alloc_table(sgt, sgt->orig_nents, GFP_KERNEL)) {
3530 		dev_err(adapter->pdev_dev, "HMA SGL allocation failed\n");
3531 		kfree(adapter->hma.sgt);
3532 		adapter->hma.sgt = NULL;
3533 		return -ENOMEM;
3534 	}
3535 
3536 	sgl = adapter->hma.sgt->sgl;
3537 	node = dev_to_node(adapter->pdev_dev);
3538 	for_each_sg(sgl, iter, sgt->orig_nents, i) {
3539 		newpage = alloc_pages_node(node, __GFP_NOWARN | GFP_KERNEL |
3540 					   __GFP_ZERO, page_order);
3541 		if (!newpage) {
3542 			dev_err(adapter->pdev_dev,
3543 				"Not enough memory for HMA page allocation\n");
3544 			ret = -ENOMEM;
3545 			goto free_hma;
3546 		}
3547 		sg_set_page(iter, newpage, page_size << page_order, 0);
3548 	}
3549 
3550 	sgt->nents = dma_map_sg(adapter->pdev_dev, sgl, sgt->orig_nents,
3551 				DMA_BIDIRECTIONAL);
3552 	if (!sgt->nents) {
3553 		dev_err(adapter->pdev_dev,
3554 			"Not enough memory for HMA DMA mapping");
3555 		ret = -ENOMEM;
3556 		goto free_hma;
3557 	}
3558 	adapter->hma.flags |= HMA_DMA_MAPPED_FLAG;
3559 
3560 	adapter->hma.phy_addr = kcalloc(sgt->nents, sizeof(dma_addr_t),
3561 					GFP_KERNEL);
3562 	if (unlikely(!adapter->hma.phy_addr))
3563 		goto free_hma;
3564 
3565 	for_each_sg(sgl, iter, sgt->nents, i) {
3566 		newpage = sg_page(iter);
3567 		adapter->hma.phy_addr[i] = sg_dma_address(iter);
3568 	}
3569 
3570 	ncmds = DIV_ROUND_UP(sgt->nents, HMA_MAX_ADDR_IN_CMD);
3571 	/* Pass on the addresses to firmware */
3572 	for (i = 0, k = 0; i < ncmds; i++, k += HMA_MAX_ADDR_IN_CMD) {
3573 		struct fw_hma_cmd hma_cmd;
3574 		u8 naddr = HMA_MAX_ADDR_IN_CMD;
3575 		u8 soc = 0, eoc = 0;
3576 		u8 hma_mode = 1; /* Presently we support only Page table mode */
3577 
3578 		soc = (i == 0) ? 1 : 0;
3579 		eoc = (i == ncmds - 1) ? 1 : 0;
3580 
3581 		/* For last cmd, set naddr corresponding to remaining
3582 		 * addresses
3583 		 */
3584 		if (i == ncmds - 1) {
3585 			naddr = sgt->nents % HMA_MAX_ADDR_IN_CMD;
3586 			naddr = naddr ? naddr : HMA_MAX_ADDR_IN_CMD;
3587 		}
3588 		memset(&hma_cmd, 0, sizeof(hma_cmd));
3589 		hma_cmd.op_pkd = htonl(FW_CMD_OP_V(FW_HMA_CMD) |
3590 				       FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3591 		hma_cmd.retval_len16 = htonl(FW_LEN16(hma_cmd));
3592 
3593 		hma_cmd.mode_to_pcie_params =
3594 			htonl(FW_HMA_CMD_MODE_V(hma_mode) |
3595 			      FW_HMA_CMD_SOC_V(soc) | FW_HMA_CMD_EOC_V(eoc));
3596 
3597 		/* HMA cmd size specified in MB's */
3598 		hma_cmd.naddr_size =
3599 			htonl(FW_HMA_CMD_SIZE_V(hma_size) |
3600 			      FW_HMA_CMD_NADDR_V(naddr));
3601 
3602 		/* Total Page size specified in units of 4K */
3603 		hma_cmd.addr_size_pkd =
3604 			htonl(FW_HMA_CMD_ADDR_SIZE_V
3605 				((page_size << page_order) >> 12));
3606 
3607 		/* Fill the 5 addresses */
3608 		for (j = 0; j < naddr; j++) {
3609 			hma_cmd.phy_address[j] =
3610 				cpu_to_be64(adapter->hma.phy_addr[j + k]);
3611 		}
3612 		ret = t4_wr_mbox(adapter, adapter->mbox, &hma_cmd,
3613 				 sizeof(hma_cmd), &hma_cmd);
3614 		if (ret) {
3615 			dev_err(adapter->pdev_dev,
3616 				"HMA FW command failed with err %d\n", ret);
3617 			goto free_hma;
3618 		}
3619 	}
3620 
3621 	if (!ret)
3622 		dev_info(adapter->pdev_dev,
3623 			 "Reserved %uMB host memory for HMA\n", hma_size);
3624 	return ret;
3625 
3626 free_hma:
3627 	adap_free_hma_mem(adapter);
3628 	return ret;
3629 }
3630 
3631 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3632 {
3633 	u32 v;
3634 	int ret;
3635 
3636 	/* Now that we've successfully configured and initialized the adapter
3637 	 * can ask the Firmware what resources it has provisioned for us.
3638 	 */
3639 	ret = t4_get_pfres(adap);
3640 	if (ret) {
3641 		dev_err(adap->pdev_dev,
3642 			"Unable to retrieve resource provisioning information\n");
3643 		return ret;
3644 	}
3645 
3646 	/* get device capabilities */
3647 	memset(c, 0, sizeof(*c));
3648 	c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3649 			       FW_CMD_REQUEST_F | FW_CMD_READ_F);
3650 	c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3651 	ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
3652 	if (ret < 0)
3653 		return ret;
3654 
3655 	c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3656 			       FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3657 	ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
3658 	if (ret < 0)
3659 		return ret;
3660 
3661 	ret = t4_config_glbl_rss(adap, adap->pf,
3662 				 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3663 				 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3664 				 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
3665 	if (ret < 0)
3666 		return ret;
3667 
3668 	ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
3669 			  MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3670 			  FW_CMD_CAP_PF);
3671 	if (ret < 0)
3672 		return ret;
3673 
3674 	t4_sge_init(adap);
3675 
3676 	/* tweak some settings */
3677 	t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3678 	t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3679 	t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3680 	v = t4_read_reg(adap, TP_PIO_DATA_A);
3681 	t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3682 
3683 	/* first 4 Tx modulation queues point to consecutive Tx channels */
3684 	adap->params.tp.tx_modq_map = 0xE4;
3685 	t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3686 		     TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3687 
3688 	/* associate each Tx modulation queue with consecutive Tx channels */
3689 	v = 0x84218421;
3690 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3691 			  &v, 1, TP_TX_SCHED_HDR_A);
3692 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3693 			  &v, 1, TP_TX_SCHED_FIFO_A);
3694 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3695 			  &v, 1, TP_TX_SCHED_PCMD_A);
3696 
3697 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3698 	if (is_offload(adap)) {
3699 		t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3700 			     TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3701 			     TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3702 			     TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3703 			     TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3704 		t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3705 			     TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3706 			     TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3707 			     TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3708 			     TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3709 	}
3710 
3711 	/* get basic stuff going */
3712 	return t4_early_init(adap, adap->pf);
3713 }
3714 
3715 /*
3716  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
3717  */
3718 #define MAX_ATIDS 8192U
3719 
3720 /*
3721  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3722  *
3723  * If the firmware we're dealing with has Configuration File support, then
3724  * we use that to perform all configuration
3725  */
3726 
3727 /*
3728  * Tweak configuration based on module parameters, etc.  Most of these have
3729  * defaults assigned to them by Firmware Configuration Files (if we're using
3730  * them) but need to be explicitly set if we're using hard-coded
3731  * initialization.  But even in the case of using Firmware Configuration
3732  * Files, we'd like to expose the ability to change these via module
3733  * parameters so these are essentially common tweaks/settings for
3734  * Configuration Files and hard-coded initialization ...
3735  */
3736 static int adap_init0_tweaks(struct adapter *adapter)
3737 {
3738 	/*
3739 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
3740 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
3741 	 * 64B Cache Line Size ...
3742 	 */
3743 	t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3744 
3745 	/*
3746 	 * Process module parameters which affect early initialization.
3747 	 */
3748 	if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3749 		dev_err(&adapter->pdev->dev,
3750 			"Ignoring illegal rx_dma_offset=%d, using 2\n",
3751 			rx_dma_offset);
3752 		rx_dma_offset = 2;
3753 	}
3754 	t4_set_reg_field(adapter, SGE_CONTROL_A,
3755 			 PKTSHIFT_V(PKTSHIFT_M),
3756 			 PKTSHIFT_V(rx_dma_offset));
3757 
3758 	/*
3759 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3760 	 * adds the pseudo header itself.
3761 	 */
3762 	t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3763 			       CSUM_HAS_PSEUDO_HDR_F, 0);
3764 
3765 	return 0;
3766 }
3767 
3768 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3769  * unto themselves and they contain their own firmware to perform their
3770  * tasks ...
3771  */
3772 static int phy_aq1202_version(const u8 *phy_fw_data,
3773 			      size_t phy_fw_size)
3774 {
3775 	int offset;
3776 
3777 	/* At offset 0x8 you're looking for the primary image's
3778 	 * starting offset which is 3 Bytes wide
3779 	 *
3780 	 * At offset 0xa of the primary image, you look for the offset
3781 	 * of the DRAM segment which is 3 Bytes wide.
3782 	 *
3783 	 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3784 	 * wide
3785 	 */
3786 	#define be16(__p) (((__p)[0] << 8) | (__p)[1])
3787 	#define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3788 	#define le24(__p) (le16(__p) | ((__p)[2] << 16))
3789 
3790 	offset = le24(phy_fw_data + 0x8) << 12;
3791 	offset = le24(phy_fw_data + offset + 0xa);
3792 	return be16(phy_fw_data + offset + 0x27e);
3793 
3794 	#undef be16
3795 	#undef le16
3796 	#undef le24
3797 }
3798 
3799 static struct info_10gbt_phy_fw {
3800 	unsigned int phy_fw_id;		/* PCI Device ID */
3801 	char *phy_fw_file;		/* /lib/firmware/ PHY Firmware file */
3802 	int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3803 	int phy_flash;			/* Has FLASH for PHY Firmware */
3804 } phy_info_array[] = {
3805 	{
3806 		PHY_AQ1202_DEVICEID,
3807 		PHY_AQ1202_FIRMWARE,
3808 		phy_aq1202_version,
3809 		1,
3810 	},
3811 	{
3812 		PHY_BCM84834_DEVICEID,
3813 		PHY_BCM84834_FIRMWARE,
3814 		NULL,
3815 		0,
3816 	},
3817 	{ 0, NULL, NULL },
3818 };
3819 
3820 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3821 {
3822 	int i;
3823 
3824 	for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3825 		if (phy_info_array[i].phy_fw_id == devid)
3826 			return &phy_info_array[i];
3827 	}
3828 	return NULL;
3829 }
3830 
3831 /* Handle updating of chip-external 10Gb/s-BT PHY firmware.  This needs to
3832  * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD.  On error
3833  * we return a negative error number.  If we transfer new firmware we return 1
3834  * (from t4_load_phy_fw()).  If we don't do anything we return 0.
3835  */
3836 static int adap_init0_phy(struct adapter *adap)
3837 {
3838 	const struct firmware *phyf;
3839 	int ret;
3840 	struct info_10gbt_phy_fw *phy_info;
3841 
3842 	/* Use the device ID to determine which PHY file to flash.
3843 	 */
3844 	phy_info = find_phy_info(adap->pdev->device);
3845 	if (!phy_info) {
3846 		dev_warn(adap->pdev_dev,
3847 			 "No PHY Firmware file found for this PHY\n");
3848 		return -EOPNOTSUPP;
3849 	}
3850 
3851 	/* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3852 	 * use that. The adapter firmware provides us with a memory buffer
3853 	 * where we can load a PHY firmware file from the host if we want to
3854 	 * override the PHY firmware File in flash.
3855 	 */
3856 	ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3857 				      adap->pdev_dev);
3858 	if (ret < 0) {
3859 		/* For adapters without FLASH attached to PHY for their
3860 		 * firmware, it's obviously a fatal error if we can't get the
3861 		 * firmware to the adapter.  For adapters with PHY firmware
3862 		 * FLASH storage, it's worth a warning if we can't find the
3863 		 * PHY Firmware but we'll neuter the error ...
3864 		 */
3865 		dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3866 			"/lib/firmware/%s, error %d\n",
3867 			phy_info->phy_fw_file, -ret);
3868 		if (phy_info->phy_flash) {
3869 			int cur_phy_fw_ver = 0;
3870 
3871 			t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3872 			dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3873 				 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3874 			ret = 0;
3875 		}
3876 
3877 		return ret;
3878 	}
3879 
3880 	/* Load PHY Firmware onto adapter.
3881 	 */
3882 	ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3883 			     phy_info->phy_fw_version,
3884 			     (u8 *)phyf->data, phyf->size);
3885 	if (ret < 0)
3886 		dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3887 			-ret);
3888 	else if (ret > 0) {
3889 		int new_phy_fw_ver = 0;
3890 
3891 		if (phy_info->phy_fw_version)
3892 			new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3893 								  phyf->size);
3894 		dev_info(adap->pdev_dev, "Successfully transferred PHY "
3895 			 "Firmware /lib/firmware/%s, version %#x\n",
3896 			 phy_info->phy_fw_file, new_phy_fw_ver);
3897 	}
3898 
3899 	release_firmware(phyf);
3900 
3901 	return ret;
3902 }
3903 
3904 /*
3905  * Attempt to initialize the adapter via a Firmware Configuration File.
3906  */
3907 static int adap_init0_config(struct adapter *adapter, int reset)
3908 {
3909 	struct fw_caps_config_cmd caps_cmd;
3910 	const struct firmware *cf;
3911 	unsigned long mtype = 0, maddr = 0;
3912 	u32 finiver, finicsum, cfcsum;
3913 	int ret;
3914 	int config_issued = 0;
3915 	char *fw_config_file, fw_config_file_path[256];
3916 	char *config_name = NULL;
3917 
3918 	/*
3919 	 * Reset device if necessary.
3920 	 */
3921 	if (reset) {
3922 		ret = t4_fw_reset(adapter, adapter->mbox,
3923 				  PIORSTMODE_F | PIORST_F);
3924 		if (ret < 0)
3925 			goto bye;
3926 	}
3927 
3928 	/* If this is a 10Gb/s-BT adapter make sure the chip-external
3929 	 * 10Gb/s-BT PHYs have up-to-date firmware.  Note that this step needs
3930 	 * to be performed after any global adapter RESET above since some
3931 	 * PHYs only have local RAM copies of the PHY firmware.
3932 	 */
3933 	if (is_10gbt_device(adapter->pdev->device)) {
3934 		ret = adap_init0_phy(adapter);
3935 		if (ret < 0)
3936 			goto bye;
3937 	}
3938 	/*
3939 	 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3940 	 * then use that.  Otherwise, use the configuration file stored
3941 	 * in the adapter flash ...
3942 	 */
3943 	switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3944 	case CHELSIO_T4:
3945 		fw_config_file = FW4_CFNAME;
3946 		break;
3947 	case CHELSIO_T5:
3948 		fw_config_file = FW5_CFNAME;
3949 		break;
3950 	case CHELSIO_T6:
3951 		fw_config_file = FW6_CFNAME;
3952 		break;
3953 	default:
3954 		dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3955 		       adapter->pdev->device);
3956 		ret = -EINVAL;
3957 		goto bye;
3958 	}
3959 
3960 	ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3961 	if (ret < 0) {
3962 		config_name = "On FLASH";
3963 		mtype = FW_MEMTYPE_CF_FLASH;
3964 		maddr = t4_flash_cfg_addr(adapter);
3965 	} else {
3966 		u32 params[7], val[7];
3967 
3968 		sprintf(fw_config_file_path,
3969 			"/lib/firmware/%s", fw_config_file);
3970 		config_name = fw_config_file_path;
3971 
3972 		if (cf->size >= FLASH_CFG_MAX_SIZE)
3973 			ret = -ENOMEM;
3974 		else {
3975 			params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3976 			     FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3977 			ret = t4_query_params(adapter, adapter->mbox,
3978 					      adapter->pf, 0, 1, params, val);
3979 			if (ret == 0) {
3980 				/*
3981 				 * For t4_memory_rw() below addresses and
3982 				 * sizes have to be in terms of multiples of 4
3983 				 * bytes.  So, if the Configuration File isn't
3984 				 * a multiple of 4 bytes in length we'll have
3985 				 * to write that out separately since we can't
3986 				 * guarantee that the bytes following the
3987 				 * residual byte in the buffer returned by
3988 				 * request_firmware() are zeroed out ...
3989 				 */
3990 				size_t resid = cf->size & 0x3;
3991 				size_t size = cf->size & ~0x3;
3992 				__be32 *data = (__be32 *)cf->data;
3993 
3994 				mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3995 				maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3996 
3997 				spin_lock(&adapter->win0_lock);
3998 				ret = t4_memory_rw(adapter, 0, mtype, maddr,
3999 						   size, data, T4_MEMORY_WRITE);
4000 				if (ret == 0 && resid != 0) {
4001 					union {
4002 						__be32 word;
4003 						char buf[4];
4004 					} last;
4005 					int i;
4006 
4007 					last.word = data[size >> 2];
4008 					for (i = resid; i < 4; i++)
4009 						last.buf[i] = 0;
4010 					ret = t4_memory_rw(adapter, 0, mtype,
4011 							   maddr + size,
4012 							   4, &last.word,
4013 							   T4_MEMORY_WRITE);
4014 				}
4015 				spin_unlock(&adapter->win0_lock);
4016 			}
4017 		}
4018 
4019 		release_firmware(cf);
4020 		if (ret)
4021 			goto bye;
4022 	}
4023 
4024 	/*
4025 	 * Issue a Capability Configuration command to the firmware to get it
4026 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
4027 	 * because we want the ability to modify various features after we've
4028 	 * processed the configuration file ...
4029 	 */
4030 	memset(&caps_cmd, 0, sizeof(caps_cmd));
4031 	caps_cmd.op_to_write =
4032 		htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4033 		      FW_CMD_REQUEST_F |
4034 		      FW_CMD_READ_F);
4035 	caps_cmd.cfvalid_to_len16 =
4036 		htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
4037 		      FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
4038 		      FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
4039 		      FW_LEN16(caps_cmd));
4040 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4041 			 &caps_cmd);
4042 
4043 	/* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
4044 	 * Configuration File in FLASH), our last gasp effort is to use the
4045 	 * Firmware Configuration File which is embedded in the firmware.  A
4046 	 * very few early versions of the firmware didn't have one embedded
4047 	 * but we can ignore those.
4048 	 */
4049 	if (ret == -ENOENT) {
4050 		memset(&caps_cmd, 0, sizeof(caps_cmd));
4051 		caps_cmd.op_to_write =
4052 			htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4053 					FW_CMD_REQUEST_F |
4054 					FW_CMD_READ_F);
4055 		caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4056 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
4057 				sizeof(caps_cmd), &caps_cmd);
4058 		config_name = "Firmware Default";
4059 	}
4060 
4061 	config_issued = 1;
4062 	if (ret < 0)
4063 		goto bye;
4064 
4065 	finiver = ntohl(caps_cmd.finiver);
4066 	finicsum = ntohl(caps_cmd.finicsum);
4067 	cfcsum = ntohl(caps_cmd.cfcsum);
4068 	if (finicsum != cfcsum)
4069 		dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4070 			 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4071 			 finicsum, cfcsum);
4072 
4073 	/*
4074 	 * And now tell the firmware to use the configuration we just loaded.
4075 	 */
4076 	caps_cmd.op_to_write =
4077 		htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4078 		      FW_CMD_REQUEST_F |
4079 		      FW_CMD_WRITE_F);
4080 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4081 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4082 			 NULL);
4083 	if (ret < 0)
4084 		goto bye;
4085 
4086 	/*
4087 	 * Tweak configuration based on system architecture, module
4088 	 * parameters, etc.
4089 	 */
4090 	ret = adap_init0_tweaks(adapter);
4091 	if (ret < 0)
4092 		goto bye;
4093 
4094 	/* We will proceed even if HMA init fails. */
4095 	ret = adap_config_hma(adapter);
4096 	if (ret)
4097 		dev_err(adapter->pdev_dev,
4098 			"HMA configuration failed with error %d\n", ret);
4099 
4100 	/*
4101 	 * And finally tell the firmware to initialize itself using the
4102 	 * parameters from the Configuration File.
4103 	 */
4104 	ret = t4_fw_initialize(adapter, adapter->mbox);
4105 	if (ret < 0)
4106 		goto bye;
4107 
4108 	/* Emit Firmware Configuration File information and return
4109 	 * successfully.
4110 	 */
4111 	dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4112 		 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4113 		 config_name, finiver, cfcsum);
4114 	return 0;
4115 
4116 	/*
4117 	 * Something bad happened.  Return the error ...  (If the "error"
4118 	 * is that there's no Configuration File on the adapter we don't
4119 	 * want to issue a warning since this is fairly common.)
4120 	 */
4121 bye:
4122 	if (config_issued && ret != -ENOENT)
4123 		dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
4124 			 config_name, -ret);
4125 	return ret;
4126 }
4127 
4128 static struct fw_info fw_info_array[] = {
4129 	{
4130 		.chip = CHELSIO_T4,
4131 		.fs_name = FW4_CFNAME,
4132 		.fw_mod_name = FW4_FNAME,
4133 		.fw_hdr = {
4134 			.chip = FW_HDR_CHIP_T4,
4135 			.fw_ver = __cpu_to_be32(FW_VERSION(T4)),
4136 			.intfver_nic = FW_INTFVER(T4, NIC),
4137 			.intfver_vnic = FW_INTFVER(T4, VNIC),
4138 			.intfver_ri = FW_INTFVER(T4, RI),
4139 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
4140 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
4141 		},
4142 	}, {
4143 		.chip = CHELSIO_T5,
4144 		.fs_name = FW5_CFNAME,
4145 		.fw_mod_name = FW5_FNAME,
4146 		.fw_hdr = {
4147 			.chip = FW_HDR_CHIP_T5,
4148 			.fw_ver = __cpu_to_be32(FW_VERSION(T5)),
4149 			.intfver_nic = FW_INTFVER(T5, NIC),
4150 			.intfver_vnic = FW_INTFVER(T5, VNIC),
4151 			.intfver_ri = FW_INTFVER(T5, RI),
4152 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
4153 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
4154 		},
4155 	}, {
4156 		.chip = CHELSIO_T6,
4157 		.fs_name = FW6_CFNAME,
4158 		.fw_mod_name = FW6_FNAME,
4159 		.fw_hdr = {
4160 			.chip = FW_HDR_CHIP_T6,
4161 			.fw_ver = __cpu_to_be32(FW_VERSION(T6)),
4162 			.intfver_nic = FW_INTFVER(T6, NIC),
4163 			.intfver_vnic = FW_INTFVER(T6, VNIC),
4164 			.intfver_ofld = FW_INTFVER(T6, OFLD),
4165 			.intfver_ri = FW_INTFVER(T6, RI),
4166 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
4167 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
4168 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
4169 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
4170 		},
4171 	}
4172 
4173 };
4174 
4175 static struct fw_info *find_fw_info(int chip)
4176 {
4177 	int i;
4178 
4179 	for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
4180 		if (fw_info_array[i].chip == chip)
4181 			return &fw_info_array[i];
4182 	}
4183 	return NULL;
4184 }
4185 
4186 /*
4187  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4188  */
4189 static int adap_init0(struct adapter *adap)
4190 {
4191 	int ret;
4192 	u32 v, port_vec;
4193 	enum dev_state state;
4194 	u32 params[7], val[7];
4195 	struct fw_caps_config_cmd caps_cmd;
4196 	int reset = 1;
4197 
4198 	/* Grab Firmware Device Log parameters as early as possible so we have
4199 	 * access to it for debugging, etc.
4200 	 */
4201 	ret = t4_init_devlog_params(adap);
4202 	if (ret < 0)
4203 		return ret;
4204 
4205 	/* Contact FW, advertising Master capability */
4206 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
4207 			  is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
4208 	if (ret < 0) {
4209 		dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4210 			ret);
4211 		return ret;
4212 	}
4213 	if (ret == adap->mbox)
4214 		adap->flags |= CXGB4_MASTER_PF;
4215 
4216 	/*
4217 	 * If we're the Master PF Driver and the device is uninitialized,
4218 	 * then let's consider upgrading the firmware ...  (We always want
4219 	 * to check the firmware version number in order to A. get it for
4220 	 * later reporting and B. to warn if the currently loaded firmware
4221 	 * is excessively mismatched relative to the driver.)
4222 	 */
4223 
4224 	t4_get_version_info(adap);
4225 	ret = t4_check_fw_version(adap);
4226 	/* If firmware is too old (not supported by driver) force an update. */
4227 	if (ret)
4228 		state = DEV_STATE_UNINIT;
4229 	if ((adap->flags & CXGB4_MASTER_PF) && state != DEV_STATE_INIT) {
4230 		struct fw_info *fw_info;
4231 		struct fw_hdr *card_fw;
4232 		const struct firmware *fw;
4233 		const u8 *fw_data = NULL;
4234 		unsigned int fw_size = 0;
4235 
4236 		/* This is the firmware whose headers the driver was compiled
4237 		 * against
4238 		 */
4239 		fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
4240 		if (fw_info == NULL) {
4241 			dev_err(adap->pdev_dev,
4242 				"unable to get firmware info for chip %d.\n",
4243 				CHELSIO_CHIP_VERSION(adap->params.chip));
4244 			return -EINVAL;
4245 		}
4246 
4247 		/* allocate memory to read the header of the firmware on the
4248 		 * card
4249 		 */
4250 		card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
4251 		if (!card_fw) {
4252 			ret = -ENOMEM;
4253 			goto bye;
4254 		}
4255 
4256 		/* Get FW from from /lib/firmware/ */
4257 		ret = request_firmware(&fw, fw_info->fw_mod_name,
4258 				       adap->pdev_dev);
4259 		if (ret < 0) {
4260 			dev_err(adap->pdev_dev,
4261 				"unable to load firmware image %s, error %d\n",
4262 				fw_info->fw_mod_name, ret);
4263 		} else {
4264 			fw_data = fw->data;
4265 			fw_size = fw->size;
4266 		}
4267 
4268 		/* upgrade FW logic */
4269 		ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
4270 				 state, &reset);
4271 
4272 		/* Cleaning up */
4273 		release_firmware(fw);
4274 		kvfree(card_fw);
4275 
4276 		if (ret < 0)
4277 			goto bye;
4278 	}
4279 
4280 	/* If the firmware is initialized already, emit a simply note to that
4281 	 * effect. Otherwise, it's time to try initializing the adapter.
4282 	 */
4283 	if (state == DEV_STATE_INIT) {
4284 		ret = adap_config_hma(adap);
4285 		if (ret)
4286 			dev_err(adap->pdev_dev,
4287 				"HMA configuration failed with error %d\n",
4288 				ret);
4289 		dev_info(adap->pdev_dev, "Coming up as %s: "\
4290 			 "Adapter already initialized\n",
4291 			 adap->flags & CXGB4_MASTER_PF ? "MASTER" : "SLAVE");
4292 	} else {
4293 		dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4294 			 "Initializing adapter\n");
4295 
4296 		/* Find out whether we're dealing with a version of the
4297 		 * firmware which has configuration file support.
4298 		 */
4299 		params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4300 			     FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4301 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
4302 				      params, val);
4303 
4304 		/* If the firmware doesn't support Configuration Files,
4305 		 * return an error.
4306 		 */
4307 		if (ret < 0) {
4308 			dev_err(adap->pdev_dev, "firmware doesn't support "
4309 				"Firmware Configuration Files\n");
4310 			goto bye;
4311 		}
4312 
4313 		/* The firmware provides us with a memory buffer where we can
4314 		 * load a Configuration File from the host if we want to
4315 		 * override the Configuration File in flash.
4316 		 */
4317 		ret = adap_init0_config(adap, reset);
4318 		if (ret == -ENOENT) {
4319 			dev_err(adap->pdev_dev, "no Configuration File "
4320 				"present on adapter.\n");
4321 			goto bye;
4322 		}
4323 		if (ret < 0) {
4324 			dev_err(adap->pdev_dev, "could not initialize "
4325 				"adapter, error %d\n", -ret);
4326 			goto bye;
4327 		}
4328 	}
4329 
4330 	/* Now that we've successfully configured and initialized the adapter
4331 	 * (or found it already initialized), we can ask the Firmware what
4332 	 * resources it has provisioned for us.
4333 	 */
4334 	ret = t4_get_pfres(adap);
4335 	if (ret) {
4336 		dev_err(adap->pdev_dev,
4337 			"Unable to retrieve resource provisioning information\n");
4338 		goto bye;
4339 	}
4340 
4341 	/* Grab VPD parameters.  This should be done after we establish a
4342 	 * connection to the firmware since some of the VPD parameters
4343 	 * (notably the Core Clock frequency) are retrieved via requests to
4344 	 * the firmware.  On the other hand, we need these fairly early on
4345 	 * so we do this right after getting ahold of the firmware.
4346 	 *
4347 	 * We need to do this after initializing the adapter because someone
4348 	 * could have FLASHed a new VPD which won't be read by the firmware
4349 	 * until we do the RESET ...
4350 	 */
4351 	ret = t4_get_vpd_params(adap, &adap->params.vpd);
4352 	if (ret < 0)
4353 		goto bye;
4354 
4355 	/* Find out what ports are available to us.  Note that we need to do
4356 	 * this before calling adap_init0_no_config() since it needs nports
4357 	 * and portvec ...
4358 	 */
4359 	v =
4360 	    FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4361 	    FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
4362 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
4363 	if (ret < 0)
4364 		goto bye;
4365 
4366 	adap->params.nports = hweight32(port_vec);
4367 	adap->params.portvec = port_vec;
4368 
4369 	/* Give the SGE code a chance to pull in anything that it needs ...
4370 	 * Note that this must be called after we retrieve our VPD parameters
4371 	 * in order to know how to convert core ticks to seconds, etc.
4372 	 */
4373 	ret = t4_sge_init(adap);
4374 	if (ret < 0)
4375 		goto bye;
4376 
4377 	/* Grab the SGE Doorbell Queue Timer values.  If successful, that
4378 	 * indicates that the Firmware and Hardware support this.
4379 	 */
4380 	params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4381 		    FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_DBQ_TIMERTICK));
4382 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4383 			      1, params, val);
4384 
4385 	if (!ret) {
4386 		adap->sge.dbqtimer_tick = val[0];
4387 		ret = t4_read_sge_dbqtimers(adap,
4388 					    ARRAY_SIZE(adap->sge.dbqtimer_val),
4389 					    adap->sge.dbqtimer_val);
4390 	}
4391 
4392 	if (!ret)
4393 		adap->flags |= CXGB4_SGE_DBQ_TIMER;
4394 
4395 	if (is_bypass_device(adap->pdev->device))
4396 		adap->params.bypass = 1;
4397 
4398 	/*
4399 	 * Grab some of our basic fundamental operating parameters.
4400 	 */
4401 #define FW_PARAM_DEV(param) \
4402 	(FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
4403 	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
4404 
4405 #define FW_PARAM_PFVF(param) \
4406 	FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
4407 	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)|  \
4408 	FW_PARAMS_PARAM_Y_V(0) | \
4409 	FW_PARAMS_PARAM_Z_V(0)
4410 
4411 	params[0] = FW_PARAM_PFVF(EQ_START);
4412 	params[1] = FW_PARAM_PFVF(L2T_START);
4413 	params[2] = FW_PARAM_PFVF(L2T_END);
4414 	params[3] = FW_PARAM_PFVF(FILTER_START);
4415 	params[4] = FW_PARAM_PFVF(FILTER_END);
4416 	params[5] = FW_PARAM_PFVF(IQFLINT_START);
4417 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
4418 	if (ret < 0)
4419 		goto bye;
4420 	adap->sge.egr_start = val[0];
4421 	adap->l2t_start = val[1];
4422 	adap->l2t_end = val[2];
4423 	adap->tids.ftid_base = val[3];
4424 	adap->tids.nftids = val[4] - val[3] + 1;
4425 	adap->sge.ingr_start = val[5];
4426 
4427 	if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
4428 		/* Read the raw mps entries. In T6, the last 2 tcam entries
4429 		 * are reserved for raw mac addresses (rawf = 2, one per port).
4430 		 */
4431 		params[0] = FW_PARAM_PFVF(RAWF_START);
4432 		params[1] = FW_PARAM_PFVF(RAWF_END);
4433 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4434 				      params, val);
4435 		if (ret == 0) {
4436 			adap->rawf_start = val[0];
4437 			adap->rawf_cnt = val[1] - val[0] + 1;
4438 		}
4439 	}
4440 
4441 	/* qids (ingress/egress) returned from firmware can be anywhere
4442 	 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
4443 	 * Hence driver needs to allocate memory for this range to
4444 	 * store the queue info. Get the highest IQFLINT/EQ index returned
4445 	 * in FW_EQ_*_CMD.alloc command.
4446 	 */
4447 	params[0] = FW_PARAM_PFVF(EQ_END);
4448 	params[1] = FW_PARAM_PFVF(IQFLINT_END);
4449 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4450 	if (ret < 0)
4451 		goto bye;
4452 	adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
4453 	adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
4454 
4455 	adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
4456 				    sizeof(*adap->sge.egr_map), GFP_KERNEL);
4457 	if (!adap->sge.egr_map) {
4458 		ret = -ENOMEM;
4459 		goto bye;
4460 	}
4461 
4462 	adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
4463 				     sizeof(*adap->sge.ingr_map), GFP_KERNEL);
4464 	if (!adap->sge.ingr_map) {
4465 		ret = -ENOMEM;
4466 		goto bye;
4467 	}
4468 
4469 	/* Allocate the memory for the vaious egress queue bitmaps
4470 	 * ie starving_fl, txq_maperr and blocked_fl.
4471 	 */
4472 	adap->sge.starving_fl =	kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4473 					sizeof(long), GFP_KERNEL);
4474 	if (!adap->sge.starving_fl) {
4475 		ret = -ENOMEM;
4476 		goto bye;
4477 	}
4478 
4479 	adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4480 				       sizeof(long), GFP_KERNEL);
4481 	if (!adap->sge.txq_maperr) {
4482 		ret = -ENOMEM;
4483 		goto bye;
4484 	}
4485 
4486 #ifdef CONFIG_DEBUG_FS
4487 	adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4488 				       sizeof(long), GFP_KERNEL);
4489 	if (!adap->sge.blocked_fl) {
4490 		ret = -ENOMEM;
4491 		goto bye;
4492 	}
4493 #endif
4494 
4495 	params[0] = FW_PARAM_PFVF(CLIP_START);
4496 	params[1] = FW_PARAM_PFVF(CLIP_END);
4497 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4498 	if (ret < 0)
4499 		goto bye;
4500 	adap->clipt_start = val[0];
4501 	adap->clipt_end = val[1];
4502 
4503 	/* We don't yet have a PARAMs calls to retrieve the number of Traffic
4504 	 * Classes supported by the hardware/firmware so we hard code it here
4505 	 * for now.
4506 	 */
4507 	adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
4508 
4509 	/* query params related to active filter region */
4510 	params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4511 	params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4512 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4513 	/* If Active filter size is set we enable establishing
4514 	 * offload connection through firmware work request
4515 	 */
4516 	if ((val[0] != val[1]) && (ret >= 0)) {
4517 		adap->flags |= CXGB4_FW_OFLD_CONN;
4518 		adap->tids.aftid_base = val[0];
4519 		adap->tids.aftid_end = val[1];
4520 	}
4521 
4522 	/* If we're running on newer firmware, let it know that we're
4523 	 * prepared to deal with encapsulated CPL messages.  Older
4524 	 * firmware won't understand this and we'll just get
4525 	 * unencapsulated messages ...
4526 	 */
4527 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4528 	val[0] = 1;
4529 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
4530 
4531 	/*
4532 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
4533 	 * capability.  Earlier versions of the firmware didn't have the
4534 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
4535 	 * permission to use ULPTX MEMWRITE DSGL.
4536 	 */
4537 	if (is_t4(adap->params.chip)) {
4538 		adap->params.ulptx_memwrite_dsgl = false;
4539 	} else {
4540 		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4541 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4542 				      1, params, val);
4543 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
4544 	}
4545 
4546 	/* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
4547 	params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
4548 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4549 			      1, params, val);
4550 	adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
4551 
4552 	/* See if FW supports FW_FILTER2 work request */
4553 	if (is_t4(adap->params.chip)) {
4554 		adap->params.filter2_wr_support = 0;
4555 	} else {
4556 		params[0] = FW_PARAM_DEV(FILTER2_WR);
4557 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4558 				      1, params, val);
4559 		adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
4560 	}
4561 
4562 	/* Check if FW supports returning vin and smt index.
4563 	 * If this is not supported, driver will interpret
4564 	 * these values from viid.
4565 	 */
4566 	params[0] = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4567 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4568 			      1, params, val);
4569 	adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
4570 
4571 	/*
4572 	 * Get device capabilities so we can determine what resources we need
4573 	 * to manage.
4574 	 */
4575 	memset(&caps_cmd, 0, sizeof(caps_cmd));
4576 	caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4577 				     FW_CMD_REQUEST_F | FW_CMD_READ_F);
4578 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4579 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4580 			 &caps_cmd);
4581 	if (ret < 0)
4582 		goto bye;
4583 
4584 	if (caps_cmd.ofldcaps ||
4585 	    (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER))) {
4586 		/* query offload-related parameters */
4587 		params[0] = FW_PARAM_DEV(NTID);
4588 		params[1] = FW_PARAM_PFVF(SERVER_START);
4589 		params[2] = FW_PARAM_PFVF(SERVER_END);
4590 		params[3] = FW_PARAM_PFVF(TDDP_START);
4591 		params[4] = FW_PARAM_PFVF(TDDP_END);
4592 		params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4593 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4594 				      params, val);
4595 		if (ret < 0)
4596 			goto bye;
4597 		adap->tids.ntids = val[0];
4598 		adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4599 		adap->tids.stid_base = val[1];
4600 		adap->tids.nstids = val[2] - val[1] + 1;
4601 		/*
4602 		 * Setup server filter region. Divide the available filter
4603 		 * region into two parts. Regular filters get 1/3rd and server
4604 		 * filters get 2/3rd part. This is only enabled if workarond
4605 		 * path is enabled.
4606 		 * 1. For regular filters.
4607 		 * 2. Server filter: This are special filters which are used
4608 		 * to redirect SYN packets to offload queue.
4609 		 */
4610 		if (adap->flags & CXGB4_FW_OFLD_CONN && !is_bypass(adap)) {
4611 			adap->tids.sftid_base = adap->tids.ftid_base +
4612 					DIV_ROUND_UP(adap->tids.nftids, 3);
4613 			adap->tids.nsftids = adap->tids.nftids -
4614 					 DIV_ROUND_UP(adap->tids.nftids, 3);
4615 			adap->tids.nftids = adap->tids.sftid_base -
4616 						adap->tids.ftid_base;
4617 		}
4618 		adap->vres.ddp.start = val[3];
4619 		adap->vres.ddp.size = val[4] - val[3] + 1;
4620 		adap->params.ofldq_wr_cred = val[5];
4621 
4622 		if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
4623 			ret = init_hash_filter(adap);
4624 			if (ret < 0)
4625 				goto bye;
4626 		} else {
4627 			adap->params.offload = 1;
4628 			adap->num_ofld_uld += 1;
4629 		}
4630 	}
4631 	if (caps_cmd.rdmacaps) {
4632 		params[0] = FW_PARAM_PFVF(STAG_START);
4633 		params[1] = FW_PARAM_PFVF(STAG_END);
4634 		params[2] = FW_PARAM_PFVF(RQ_START);
4635 		params[3] = FW_PARAM_PFVF(RQ_END);
4636 		params[4] = FW_PARAM_PFVF(PBL_START);
4637 		params[5] = FW_PARAM_PFVF(PBL_END);
4638 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4639 				      params, val);
4640 		if (ret < 0)
4641 			goto bye;
4642 		adap->vres.stag.start = val[0];
4643 		adap->vres.stag.size = val[1] - val[0] + 1;
4644 		adap->vres.rq.start = val[2];
4645 		adap->vres.rq.size = val[3] - val[2] + 1;
4646 		adap->vres.pbl.start = val[4];
4647 		adap->vres.pbl.size = val[5] - val[4] + 1;
4648 
4649 		params[0] = FW_PARAM_PFVF(SRQ_START);
4650 		params[1] = FW_PARAM_PFVF(SRQ_END);
4651 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4652 				      params, val);
4653 		if (!ret) {
4654 			adap->vres.srq.start = val[0];
4655 			adap->vres.srq.size = val[1] - val[0] + 1;
4656 		}
4657 		if (adap->vres.srq.size) {
4658 			adap->srq = t4_init_srq(adap->vres.srq.size);
4659 			if (!adap->srq)
4660 				dev_warn(&adap->pdev->dev, "could not allocate SRQ, continuing\n");
4661 		}
4662 
4663 		params[0] = FW_PARAM_PFVF(SQRQ_START);
4664 		params[1] = FW_PARAM_PFVF(SQRQ_END);
4665 		params[2] = FW_PARAM_PFVF(CQ_START);
4666 		params[3] = FW_PARAM_PFVF(CQ_END);
4667 		params[4] = FW_PARAM_PFVF(OCQ_START);
4668 		params[5] = FW_PARAM_PFVF(OCQ_END);
4669 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
4670 				      val);
4671 		if (ret < 0)
4672 			goto bye;
4673 		adap->vres.qp.start = val[0];
4674 		adap->vres.qp.size = val[1] - val[0] + 1;
4675 		adap->vres.cq.start = val[2];
4676 		adap->vres.cq.size = val[3] - val[2] + 1;
4677 		adap->vres.ocq.start = val[4];
4678 		adap->vres.ocq.size = val[5] - val[4] + 1;
4679 
4680 		params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
4681 		params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4682 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
4683 				      val);
4684 		if (ret < 0) {
4685 			adap->params.max_ordird_qp = 8;
4686 			adap->params.max_ird_adapter = 32 * adap->tids.ntids;
4687 			ret = 0;
4688 		} else {
4689 			adap->params.max_ordird_qp = val[0];
4690 			adap->params.max_ird_adapter = val[1];
4691 		}
4692 		dev_info(adap->pdev_dev,
4693 			 "max_ordird_qp %d max_ird_adapter %d\n",
4694 			 adap->params.max_ordird_qp,
4695 			 adap->params.max_ird_adapter);
4696 
4697 		/* Enable write_with_immediate if FW supports it */
4698 		params[0] = FW_PARAM_DEV(RDMA_WRITE_WITH_IMM);
4699 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
4700 				      val);
4701 		adap->params.write_w_imm_support = (ret == 0 && val[0] != 0);
4702 
4703 		/* Enable write_cmpl if FW supports it */
4704 		params[0] = FW_PARAM_DEV(RI_WRITE_CMPL_WR);
4705 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
4706 				      val);
4707 		adap->params.write_cmpl_support = (ret == 0 && val[0] != 0);
4708 		adap->num_ofld_uld += 2;
4709 	}
4710 	if (caps_cmd.iscsicaps) {
4711 		params[0] = FW_PARAM_PFVF(ISCSI_START);
4712 		params[1] = FW_PARAM_PFVF(ISCSI_END);
4713 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4714 				      params, val);
4715 		if (ret < 0)
4716 			goto bye;
4717 		adap->vres.iscsi.start = val[0];
4718 		adap->vres.iscsi.size = val[1] - val[0] + 1;
4719 		/* LIO target and cxgb4i initiaitor */
4720 		adap->num_ofld_uld += 2;
4721 	}
4722 	if (caps_cmd.cryptocaps) {
4723 		if (ntohs(caps_cmd.cryptocaps) &
4724 		    FW_CAPS_CONFIG_CRYPTO_LOOKASIDE) {
4725 			params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
4726 			ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4727 					      2, params, val);
4728 			if (ret < 0) {
4729 				if (ret != -EINVAL)
4730 					goto bye;
4731 			} else {
4732 				adap->vres.ncrypto_fc = val[0];
4733 			}
4734 			adap->num_ofld_uld += 1;
4735 		}
4736 		if (ntohs(caps_cmd.cryptocaps) &
4737 		    FW_CAPS_CONFIG_TLS_INLINE) {
4738 			params[0] = FW_PARAM_PFVF(TLS_START);
4739 			params[1] = FW_PARAM_PFVF(TLS_END);
4740 			ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4741 					      2, params, val);
4742 			if (ret < 0)
4743 				goto bye;
4744 			adap->vres.key.start = val[0];
4745 			adap->vres.key.size = val[1] - val[0] + 1;
4746 			adap->num_uld += 1;
4747 		}
4748 		adap->params.crypto = ntohs(caps_cmd.cryptocaps);
4749 	}
4750 #undef FW_PARAM_PFVF
4751 #undef FW_PARAM_DEV
4752 
4753 	/* The MTU/MSS Table is initialized by now, so load their values.  If
4754 	 * we're initializing the adapter, then we'll make any modifications
4755 	 * we want to the MTU/MSS Table and also initialize the congestion
4756 	 * parameters.
4757 	 */
4758 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4759 	if (state != DEV_STATE_INIT) {
4760 		int i;
4761 
4762 		/* The default MTU Table contains values 1492 and 1500.
4763 		 * However, for TCP, it's better to have two values which are
4764 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
4765 		 * This allows us to have a TCP Data Payload which is a
4766 		 * multiple of 8 regardless of what combination of TCP Options
4767 		 * are in use (always a multiple of 4 bytes) which is
4768 		 * important for performance reasons.  For instance, if no
4769 		 * options are in use, then we have a 20-byte IP header and a
4770 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
4771 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
4772 		 * which is not a multiple of 8.  So using an MSS of 1488 in
4773 		 * this case results in a TCP Data Payload of 1448 bytes which
4774 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
4775 		 * Stamps have been negotiated, then an MTU of 1500 bytes
4776 		 * results in a TCP Data Payload of 1448 bytes which, as
4777 		 * above, is a multiple of 8 bytes ...
4778 		 */
4779 		for (i = 0; i < NMTUS; i++)
4780 			if (adap->params.mtus[i] == 1492) {
4781 				adap->params.mtus[i] = 1488;
4782 				break;
4783 			}
4784 
4785 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4786 			     adap->params.b_wnd);
4787 	}
4788 	t4_init_sge_params(adap);
4789 	adap->flags |= CXGB4_FW_OK;
4790 	t4_init_tp_params(adap, true);
4791 	return 0;
4792 
4793 	/*
4794 	 * Something bad happened.  If a command timed out or failed with EIO
4795 	 * FW does not operate within its spec or something catastrophic
4796 	 * happened to HW/FW, stop issuing commands.
4797 	 */
4798 bye:
4799 	adap_free_hma_mem(adap);
4800 	kfree(adap->sge.egr_map);
4801 	kfree(adap->sge.ingr_map);
4802 	kfree(adap->sge.starving_fl);
4803 	kfree(adap->sge.txq_maperr);
4804 #ifdef CONFIG_DEBUG_FS
4805 	kfree(adap->sge.blocked_fl);
4806 #endif
4807 	if (ret != -ETIMEDOUT && ret != -EIO)
4808 		t4_fw_bye(adap, adap->mbox);
4809 	return ret;
4810 }
4811 
4812 /* EEH callbacks */
4813 
4814 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4815 					 pci_channel_state_t state)
4816 {
4817 	int i;
4818 	struct adapter *adap = pci_get_drvdata(pdev);
4819 
4820 	if (!adap)
4821 		goto out;
4822 
4823 	rtnl_lock();
4824 	adap->flags &= ~CXGB4_FW_OK;
4825 	notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4826 	spin_lock(&adap->stats_lock);
4827 	for_each_port(adap, i) {
4828 		struct net_device *dev = adap->port[i];
4829 		if (dev) {
4830 			netif_device_detach(dev);
4831 			netif_carrier_off(dev);
4832 		}
4833 	}
4834 	spin_unlock(&adap->stats_lock);
4835 	disable_interrupts(adap);
4836 	if (adap->flags & CXGB4_FULL_INIT_DONE)
4837 		cxgb_down(adap);
4838 	rtnl_unlock();
4839 	if ((adap->flags & CXGB4_DEV_ENABLED)) {
4840 		pci_disable_device(pdev);
4841 		adap->flags &= ~CXGB4_DEV_ENABLED;
4842 	}
4843 out:	return state == pci_channel_io_perm_failure ?
4844 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4845 }
4846 
4847 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4848 {
4849 	int i, ret;
4850 	struct fw_caps_config_cmd c;
4851 	struct adapter *adap = pci_get_drvdata(pdev);
4852 
4853 	if (!adap) {
4854 		pci_restore_state(pdev);
4855 		pci_save_state(pdev);
4856 		return PCI_ERS_RESULT_RECOVERED;
4857 	}
4858 
4859 	if (!(adap->flags & CXGB4_DEV_ENABLED)) {
4860 		if (pci_enable_device(pdev)) {
4861 			dev_err(&pdev->dev, "Cannot reenable PCI "
4862 					    "device after reset\n");
4863 			return PCI_ERS_RESULT_DISCONNECT;
4864 		}
4865 		adap->flags |= CXGB4_DEV_ENABLED;
4866 	}
4867 
4868 	pci_set_master(pdev);
4869 	pci_restore_state(pdev);
4870 	pci_save_state(pdev);
4871 
4872 	if (t4_wait_dev_ready(adap->regs) < 0)
4873 		return PCI_ERS_RESULT_DISCONNECT;
4874 	if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
4875 		return PCI_ERS_RESULT_DISCONNECT;
4876 	adap->flags |= CXGB4_FW_OK;
4877 	if (adap_init1(adap, &c))
4878 		return PCI_ERS_RESULT_DISCONNECT;
4879 
4880 	for_each_port(adap, i) {
4881 		struct port_info *pi = adap2pinfo(adap, i);
4882 		u8 vivld = 0, vin = 0;
4883 
4884 		ret = t4_alloc_vi(adap, adap->mbox, pi->tx_chan, adap->pf, 0, 1,
4885 				  NULL, NULL, &vivld, &vin);
4886 		if (ret < 0)
4887 			return PCI_ERS_RESULT_DISCONNECT;
4888 		pi->viid = ret;
4889 		pi->xact_addr_filt = -1;
4890 		/* If fw supports returning the VIN as part of FW_VI_CMD,
4891 		 * save the returned values.
4892 		 */
4893 		if (adap->params.viid_smt_extn_support) {
4894 			pi->vivld = vivld;
4895 			pi->vin = vin;
4896 		} else {
4897 			/* Retrieve the values from VIID */
4898 			pi->vivld = FW_VIID_VIVLD_G(pi->viid);
4899 			pi->vin = FW_VIID_VIN_G(pi->viid);
4900 		}
4901 	}
4902 
4903 	t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4904 		     adap->params.b_wnd);
4905 	setup_memwin(adap);
4906 	if (cxgb_up(adap))
4907 		return PCI_ERS_RESULT_DISCONNECT;
4908 	return PCI_ERS_RESULT_RECOVERED;
4909 }
4910 
4911 static void eeh_resume(struct pci_dev *pdev)
4912 {
4913 	int i;
4914 	struct adapter *adap = pci_get_drvdata(pdev);
4915 
4916 	if (!adap)
4917 		return;
4918 
4919 	rtnl_lock();
4920 	for_each_port(adap, i) {
4921 		struct net_device *dev = adap->port[i];
4922 		if (dev) {
4923 			if (netif_running(dev)) {
4924 				link_start(dev);
4925 				cxgb_set_rxmode(dev);
4926 			}
4927 			netif_device_attach(dev);
4928 		}
4929 	}
4930 	rtnl_unlock();
4931 }
4932 
4933 static const struct pci_error_handlers cxgb4_eeh = {
4934 	.error_detected = eeh_err_detected,
4935 	.slot_reset     = eeh_slot_reset,
4936 	.resume         = eeh_resume,
4937 };
4938 
4939 /* Return true if the Link Configuration supports "High Speeds" (those greater
4940  * than 1Gb/s).
4941  */
4942 static inline bool is_x_10g_port(const struct link_config *lc)
4943 {
4944 	unsigned int speeds, high_speeds;
4945 
4946 	speeds = FW_PORT_CAP32_SPEED_V(FW_PORT_CAP32_SPEED_G(lc->pcaps));
4947 	high_speeds = speeds &
4948 			~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
4949 
4950 	return high_speeds != 0;
4951 }
4952 
4953 /*
4954  * Perform default configuration of DMA queues depending on the number and type
4955  * of ports we found and the number of available CPUs.  Most settings can be
4956  * modified by the admin prior to actual use.
4957  */
4958 static int cfg_queues(struct adapter *adap)
4959 {
4960 	struct sge *s = &adap->sge;
4961 	int i, n10g = 0, qidx = 0;
4962 	int niqflint, neq, avail_eth_qsets;
4963 	int max_eth_qsets = 32;
4964 #ifndef CONFIG_CHELSIO_T4_DCB
4965 	int q10g = 0;
4966 #endif
4967 
4968 	/* Reduce memory usage in kdump environment, disable all offload.
4969 	 */
4970 	if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
4971 		adap->params.offload = 0;
4972 		adap->params.crypto = 0;
4973 	}
4974 
4975 	/* Calculate the number of Ethernet Queue Sets available based on
4976 	 * resources provisioned for us.  We always have an Asynchronous
4977 	 * Firmware Event Ingress Queue.  If we're operating in MSI or Legacy
4978 	 * IRQ Pin Interrupt mode, then we'll also have a Forwarded Interrupt
4979 	 * Ingress Queue.  Meanwhile, we need two Egress Queues for each
4980 	 * Queue Set: one for the Free List and one for the Ethernet TX Queue.
4981 	 *
4982 	 * Note that we should also take into account all of the various
4983 	 * Offload Queues.  But, in any situation where we're operating in
4984 	 * a Resource Constrained Provisioning environment, doing any Offload
4985 	 * at all is problematic ...
4986 	 */
4987 	niqflint = adap->params.pfres.niqflint - 1;
4988 	if (!(adap->flags & CXGB4_USING_MSIX))
4989 		niqflint--;
4990 	neq = adap->params.pfres.neq / 2;
4991 	avail_eth_qsets = min(niqflint, neq);
4992 
4993 	if (avail_eth_qsets > max_eth_qsets)
4994 		avail_eth_qsets = max_eth_qsets;
4995 
4996 	if (avail_eth_qsets < adap->params.nports) {
4997 		dev_err(adap->pdev_dev, "avail_eth_qsets=%d < nports=%d\n",
4998 			avail_eth_qsets, adap->params.nports);
4999 		return -ENOMEM;
5000 	}
5001 
5002 	/* Count the number of 10Gb/s or better ports */
5003 	for_each_port(adap, i)
5004 		n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
5005 
5006 #ifdef CONFIG_CHELSIO_T4_DCB
5007 	/* For Data Center Bridging support we need to be able to support up
5008 	 * to 8 Traffic Priorities; each of which will be assigned to its
5009 	 * own TX Queue in order to prevent Head-Of-Line Blocking.
5010 	 */
5011 	if (adap->params.nports * 8 > avail_eth_qsets) {
5012 		dev_err(adap->pdev_dev, "DCB avail_eth_qsets=%d < %d!\n",
5013 			avail_eth_qsets, adap->params.nports * 8);
5014 		return -ENOMEM;
5015 	}
5016 
5017 	for_each_port(adap, i) {
5018 		struct port_info *pi = adap2pinfo(adap, i);
5019 
5020 		pi->first_qset = qidx;
5021 		pi->nqsets = is_kdump_kernel() ? 1 : 8;
5022 		qidx += pi->nqsets;
5023 	}
5024 #else /* !CONFIG_CHELSIO_T4_DCB */
5025 	/*
5026 	 * We default to 1 queue per non-10G port and up to # of cores queues
5027 	 * per 10G port.
5028 	 */
5029 	if (n10g)
5030 		q10g = (avail_eth_qsets - (adap->params.nports - n10g)) / n10g;
5031 	if (q10g > netif_get_num_default_rss_queues())
5032 		q10g = netif_get_num_default_rss_queues();
5033 
5034 	if (is_kdump_kernel())
5035 		q10g = 1;
5036 
5037 	for_each_port(adap, i) {
5038 		struct port_info *pi = adap2pinfo(adap, i);
5039 
5040 		pi->first_qset = qidx;
5041 		pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
5042 		qidx += pi->nqsets;
5043 	}
5044 #endif /* !CONFIG_CHELSIO_T4_DCB */
5045 
5046 	s->ethqsets = qidx;
5047 	s->max_ethqsets = qidx;   /* MSI-X may lower it later */
5048 
5049 	if (is_uld(adap)) {
5050 		/*
5051 		 * For offload we use 1 queue/channel if all ports are up to 1G,
5052 		 * otherwise we divide all available queues amongst the channels
5053 		 * capped by the number of available cores.
5054 		 */
5055 		if (n10g) {
5056 			i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
5057 			s->ofldqsets = roundup(i, adap->params.nports);
5058 		} else {
5059 			s->ofldqsets = adap->params.nports;
5060 		}
5061 	}
5062 
5063 	for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5064 		struct sge_eth_rxq *r = &s->ethrxq[i];
5065 
5066 		init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
5067 		r->fl.size = 72;
5068 	}
5069 
5070 	for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5071 		s->ethtxq[i].q.size = 1024;
5072 
5073 	for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5074 		s->ctrlq[i].q.size = 512;
5075 
5076 	if (!is_t4(adap->params.chip))
5077 		s->ptptxq.q.size = 8;
5078 
5079 	init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
5080 	init_rspq(adap, &s->intrq, 0, 1, 512, 64);
5081 
5082 	return 0;
5083 }
5084 
5085 /*
5086  * Reduce the number of Ethernet queues across all ports to at most n.
5087  * n provides at least one queue per port.
5088  */
5089 static void reduce_ethqs(struct adapter *adap, int n)
5090 {
5091 	int i;
5092 	struct port_info *pi;
5093 
5094 	while (n < adap->sge.ethqsets)
5095 		for_each_port(adap, i) {
5096 			pi = adap2pinfo(adap, i);
5097 			if (pi->nqsets > 1) {
5098 				pi->nqsets--;
5099 				adap->sge.ethqsets--;
5100 				if (adap->sge.ethqsets <= n)
5101 					break;
5102 			}
5103 		}
5104 
5105 	n = 0;
5106 	for_each_port(adap, i) {
5107 		pi = adap2pinfo(adap, i);
5108 		pi->first_qset = n;
5109 		n += pi->nqsets;
5110 	}
5111 }
5112 
5113 static int get_msix_info(struct adapter *adap)
5114 {
5115 	struct uld_msix_info *msix_info;
5116 	unsigned int max_ingq = 0;
5117 
5118 	if (is_offload(adap))
5119 		max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
5120 	if (is_pci_uld(adap))
5121 		max_ingq += MAX_OFLD_QSETS * adap->num_uld;
5122 
5123 	if (!max_ingq)
5124 		goto out;
5125 
5126 	msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
5127 	if (!msix_info)
5128 		return -ENOMEM;
5129 
5130 	adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
5131 						 sizeof(long), GFP_KERNEL);
5132 	if (!adap->msix_bmap_ulds.msix_bmap) {
5133 		kfree(msix_info);
5134 		return -ENOMEM;
5135 	}
5136 	spin_lock_init(&adap->msix_bmap_ulds.lock);
5137 	adap->msix_info_ulds = msix_info;
5138 out:
5139 	return 0;
5140 }
5141 
5142 static void free_msix_info(struct adapter *adap)
5143 {
5144 	if (!(adap->num_uld && adap->num_ofld_uld))
5145 		return;
5146 
5147 	kfree(adap->msix_info_ulds);
5148 	kfree(adap->msix_bmap_ulds.msix_bmap);
5149 }
5150 
5151 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5152 #define EXTRA_VECS 2
5153 
5154 static int enable_msix(struct adapter *adap)
5155 {
5156 	int ofld_need = 0, uld_need = 0;
5157 	int i, j, want, need, allocated;
5158 	struct sge *s = &adap->sge;
5159 	unsigned int nchan = adap->params.nports;
5160 	struct msix_entry *entries;
5161 	int max_ingq = MAX_INGQ;
5162 
5163 	if (is_pci_uld(adap))
5164 		max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
5165 	if (is_offload(adap))
5166 		max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
5167 	entries = kmalloc_array(max_ingq + 1, sizeof(*entries),
5168 				GFP_KERNEL);
5169 	if (!entries)
5170 		return -ENOMEM;
5171 
5172 	/* map for msix */
5173 	if (get_msix_info(adap)) {
5174 		adap->params.offload = 0;
5175 		adap->params.crypto = 0;
5176 	}
5177 
5178 	for (i = 0; i < max_ingq + 1; ++i)
5179 		entries[i].entry = i;
5180 
5181 	want = s->max_ethqsets + EXTRA_VECS;
5182 	if (is_offload(adap)) {
5183 		want += adap->num_ofld_uld * s->ofldqsets;
5184 		ofld_need = adap->num_ofld_uld * nchan;
5185 	}
5186 	if (is_pci_uld(adap)) {
5187 		want += adap->num_uld * s->ofldqsets;
5188 		uld_need = adap->num_uld * nchan;
5189 	}
5190 #ifdef CONFIG_CHELSIO_T4_DCB
5191 	/* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
5192 	 * each port.
5193 	 */
5194 	need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
5195 #else
5196 	need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
5197 #endif
5198 	allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
5199 	if (allocated < 0) {
5200 		dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
5201 			 " not using MSI-X\n");
5202 		kfree(entries);
5203 		return allocated;
5204 	}
5205 
5206 	/* Distribute available vectors to the various queue groups.
5207 	 * Every group gets its minimum requirement and NIC gets top
5208 	 * priority for leftovers.
5209 	 */
5210 	i = allocated - EXTRA_VECS - ofld_need - uld_need;
5211 	if (i < s->max_ethqsets) {
5212 		s->max_ethqsets = i;
5213 		if (i < s->ethqsets)
5214 			reduce_ethqs(adap, i);
5215 	}
5216 	if (is_uld(adap)) {
5217 		if (allocated < want)
5218 			s->nqs_per_uld = nchan;
5219 		else
5220 			s->nqs_per_uld = s->ofldqsets;
5221 	}
5222 
5223 	for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
5224 		adap->msix_info[i].vec = entries[i].vector;
5225 	if (is_uld(adap)) {
5226 		for (j = 0 ; i < allocated; ++i, j++) {
5227 			adap->msix_info_ulds[j].vec = entries[i].vector;
5228 			adap->msix_info_ulds[j].idx = i;
5229 		}
5230 		adap->msix_bmap_ulds.mapsize = j;
5231 	}
5232 	dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
5233 		 "nic %d per uld %d\n",
5234 		 allocated, s->max_ethqsets, s->nqs_per_uld);
5235 
5236 	kfree(entries);
5237 	return 0;
5238 }
5239 
5240 #undef EXTRA_VECS
5241 
5242 static int init_rss(struct adapter *adap)
5243 {
5244 	unsigned int i;
5245 	int err;
5246 
5247 	err = t4_init_rss_mode(adap, adap->mbox);
5248 	if (err)
5249 		return err;
5250 
5251 	for_each_port(adap, i) {
5252 		struct port_info *pi = adap2pinfo(adap, i);
5253 
5254 		pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5255 		if (!pi->rss)
5256 			return -ENOMEM;
5257 	}
5258 	return 0;
5259 }
5260 
5261 /* Dump basic information about the adapter */
5262 static void print_adapter_info(struct adapter *adapter)
5263 {
5264 	/* Hardware/Firmware/etc. Version/Revision IDs */
5265 	t4_dump_version_info(adapter);
5266 
5267 	/* Software/Hardware configuration */
5268 	dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
5269 		 is_offload(adapter) ? "R" : "",
5270 		 ((adapter->flags & CXGB4_USING_MSIX) ? "MSI-X" :
5271 		  (adapter->flags & CXGB4_USING_MSI) ? "MSI" : ""),
5272 		 is_offload(adapter) ? "Offload" : "non-Offload");
5273 }
5274 
5275 static void print_port_info(const struct net_device *dev)
5276 {
5277 	char buf[80];
5278 	char *bufp = buf;
5279 	const struct port_info *pi = netdev_priv(dev);
5280 	const struct adapter *adap = pi->adapter;
5281 
5282 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
5283 		bufp += sprintf(bufp, "100M/");
5284 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
5285 		bufp += sprintf(bufp, "1G/");
5286 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
5287 		bufp += sprintf(bufp, "10G/");
5288 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
5289 		bufp += sprintf(bufp, "25G/");
5290 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
5291 		bufp += sprintf(bufp, "40G/");
5292 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
5293 		bufp += sprintf(bufp, "50G/");
5294 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
5295 		bufp += sprintf(bufp, "100G/");
5296 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_200G)
5297 		bufp += sprintf(bufp, "200G/");
5298 	if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_400G)
5299 		bufp += sprintf(bufp, "400G/");
5300 	if (bufp != buf)
5301 		--bufp;
5302 	sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
5303 
5304 	netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
5305 		    dev->name, adap->params.vpd.id, adap->name, buf);
5306 }
5307 
5308 /*
5309  * Free the following resources:
5310  * - memory used for tables
5311  * - MSI/MSI-X
5312  * - net devices
5313  * - resources FW is holding for us
5314  */
5315 static void free_some_resources(struct adapter *adapter)
5316 {
5317 	unsigned int i;
5318 
5319 	kvfree(adapter->mps_encap);
5320 	kvfree(adapter->smt);
5321 	kvfree(adapter->l2t);
5322 	kvfree(adapter->srq);
5323 	t4_cleanup_sched(adapter);
5324 	kvfree(adapter->tids.tid_tab);
5325 	cxgb4_cleanup_tc_flower(adapter);
5326 	cxgb4_cleanup_tc_u32(adapter);
5327 	kfree(adapter->sge.egr_map);
5328 	kfree(adapter->sge.ingr_map);
5329 	kfree(adapter->sge.starving_fl);
5330 	kfree(adapter->sge.txq_maperr);
5331 #ifdef CONFIG_DEBUG_FS
5332 	kfree(adapter->sge.blocked_fl);
5333 #endif
5334 	disable_msi(adapter);
5335 
5336 	for_each_port(adapter, i)
5337 		if (adapter->port[i]) {
5338 			struct port_info *pi = adap2pinfo(adapter, i);
5339 
5340 			if (pi->viid != 0)
5341 				t4_free_vi(adapter, adapter->mbox, adapter->pf,
5342 					   0, pi->viid);
5343 			kfree(adap2pinfo(adapter, i)->rss);
5344 			free_netdev(adapter->port[i]);
5345 		}
5346 	if (adapter->flags & CXGB4_FW_OK)
5347 		t4_fw_bye(adapter, adapter->pf);
5348 }
5349 
5350 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5351 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5352 		   NETIF_F_GRO | NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5353 #define SEGMENT_SIZE 128
5354 
5355 static int t4_get_chip_type(struct adapter *adap, int ver)
5356 {
5357 	u32 pl_rev = REV_G(t4_read_reg(adap, PL_REV_A));
5358 
5359 	switch (ver) {
5360 	case CHELSIO_T4:
5361 		return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
5362 	case CHELSIO_T5:
5363 		return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
5364 	case CHELSIO_T6:
5365 		return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
5366 	default:
5367 		break;
5368 	}
5369 	return -EINVAL;
5370 }
5371 
5372 #ifdef CONFIG_PCI_IOV
5373 static void cxgb4_mgmt_setup(struct net_device *dev)
5374 {
5375 	dev->type = ARPHRD_NONE;
5376 	dev->mtu = 0;
5377 	dev->hard_header_len = 0;
5378 	dev->addr_len = 0;
5379 	dev->tx_queue_len = 0;
5380 	dev->flags |= IFF_NOARP;
5381 	dev->priv_flags |= IFF_NO_QUEUE;
5382 
5383 	/* Initialize the device structure. */
5384 	dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
5385 	dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
5386 }
5387 
5388 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
5389 {
5390 	struct adapter *adap = pci_get_drvdata(pdev);
5391 	int err = 0;
5392 	int current_vfs = pci_num_vf(pdev);
5393 	u32 pcie_fw;
5394 
5395 	pcie_fw = readl(adap->regs + PCIE_FW_A);
5396 	/* Check if fw is initialized */
5397 	if (!(pcie_fw & PCIE_FW_INIT_F)) {
5398 		dev_warn(&pdev->dev, "Device not initialized\n");
5399 		return -EOPNOTSUPP;
5400 	}
5401 
5402 	/* If any of the VF's is already assigned to Guest OS, then
5403 	 * SRIOV for the same cannot be modified
5404 	 */
5405 	if (current_vfs && pci_vfs_assigned(pdev)) {
5406 		dev_err(&pdev->dev,
5407 			"Cannot modify SR-IOV while VFs are assigned\n");
5408 		return current_vfs;
5409 	}
5410 	/* Note that the upper-level code ensures that we're never called with
5411 	 * a non-zero "num_vfs" when we already have VFs instantiated.  But
5412 	 * it never hurts to code defensively.
5413 	 */
5414 	if (num_vfs != 0 && current_vfs != 0)
5415 		return -EBUSY;
5416 
5417 	/* Nothing to do for no change. */
5418 	if (num_vfs == current_vfs)
5419 		return num_vfs;
5420 
5421 	/* Disable SRIOV when zero is passed. */
5422 	if (!num_vfs) {
5423 		pci_disable_sriov(pdev);
5424 		/* free VF Management Interface */
5425 		unregister_netdev(adap->port[0]);
5426 		free_netdev(adap->port[0]);
5427 		adap->port[0] = NULL;
5428 
5429 		/* free VF resources */
5430 		adap->num_vfs = 0;
5431 		kfree(adap->vfinfo);
5432 		adap->vfinfo = NULL;
5433 		return 0;
5434 	}
5435 
5436 	if (!current_vfs) {
5437 		struct fw_pfvf_cmd port_cmd, port_rpl;
5438 		struct net_device *netdev;
5439 		unsigned int pmask, port;
5440 		struct pci_dev *pbridge;
5441 		struct port_info *pi;
5442 		char name[IFNAMSIZ];
5443 		u32 devcap2;
5444 		u16 flags;
5445 		int pos;
5446 
5447 		/* If we want to instantiate Virtual Functions, then our
5448 		 * parent bridge's PCI-E needs to support Alternative Routing
5449 		 * ID (ARI) because our VFs will show up at function offset 8
5450 		 * and above.
5451 		 */
5452 		pbridge = pdev->bus->self;
5453 		pos = pci_find_capability(pbridge, PCI_CAP_ID_EXP);
5454 		pci_read_config_word(pbridge, pos + PCI_EXP_FLAGS, &flags);
5455 		pci_read_config_dword(pbridge, pos + PCI_EXP_DEVCAP2, &devcap2);
5456 
5457 		if ((flags & PCI_EXP_FLAGS_VERS) < 2 ||
5458 		    !(devcap2 & PCI_EXP_DEVCAP2_ARI)) {
5459 			/* Our parent bridge does not support ARI so issue a
5460 			 * warning and skip instantiating the VFs.  They
5461 			 * won't be reachable.
5462 			 */
5463 			dev_warn(&pdev->dev, "Parent bridge %02x:%02x.%x doesn't support ARI; can't instantiate Virtual Functions\n",
5464 				 pbridge->bus->number, PCI_SLOT(pbridge->devfn),
5465 				 PCI_FUNC(pbridge->devfn));
5466 			return -ENOTSUPP;
5467 		}
5468 		memset(&port_cmd, 0, sizeof(port_cmd));
5469 		port_cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_PFVF_CMD) |
5470 						 FW_CMD_REQUEST_F |
5471 						 FW_CMD_READ_F |
5472 						 FW_PFVF_CMD_PFN_V(adap->pf) |
5473 						 FW_PFVF_CMD_VFN_V(0));
5474 		port_cmd.retval_len16 = cpu_to_be32(FW_LEN16(port_cmd));
5475 		err = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
5476 				 &port_rpl);
5477 		if (err)
5478 			return err;
5479 		pmask = FW_PFVF_CMD_PMASK_G(be32_to_cpu(port_rpl.type_to_neq));
5480 		port = ffs(pmask) - 1;
5481 		/* Allocate VF Management Interface. */
5482 		snprintf(name, IFNAMSIZ, "mgmtpf%d,%d", adap->adap_idx,
5483 			 adap->pf);
5484 		netdev = alloc_netdev(sizeof(struct port_info),
5485 				      name, NET_NAME_UNKNOWN, cxgb4_mgmt_setup);
5486 		if (!netdev)
5487 			return -ENOMEM;
5488 
5489 		pi = netdev_priv(netdev);
5490 		pi->adapter = adap;
5491 		pi->lport = port;
5492 		pi->tx_chan = port;
5493 		SET_NETDEV_DEV(netdev, &pdev->dev);
5494 
5495 		adap->port[0] = netdev;
5496 		pi->port_id = 0;
5497 
5498 		err = register_netdev(adap->port[0]);
5499 		if (err) {
5500 			pr_info("Unable to register VF mgmt netdev %s\n", name);
5501 			free_netdev(adap->port[0]);
5502 			adap->port[0] = NULL;
5503 			return err;
5504 		}
5505 		/* Allocate and set up VF Information. */
5506 		adap->vfinfo = kcalloc(pci_sriov_get_totalvfs(pdev),
5507 				       sizeof(struct vf_info), GFP_KERNEL);
5508 		if (!adap->vfinfo) {
5509 			unregister_netdev(adap->port[0]);
5510 			free_netdev(adap->port[0]);
5511 			adap->port[0] = NULL;
5512 			return -ENOMEM;
5513 		}
5514 		cxgb4_mgmt_fill_vf_station_mac_addr(adap);
5515 	}
5516 	/* Instantiate the requested number of VFs. */
5517 	err = pci_enable_sriov(pdev, num_vfs);
5518 	if (err) {
5519 		pr_info("Unable to instantiate %d VFs\n", num_vfs);
5520 		if (!current_vfs) {
5521 			unregister_netdev(adap->port[0]);
5522 			free_netdev(adap->port[0]);
5523 			adap->port[0] = NULL;
5524 			kfree(adap->vfinfo);
5525 			adap->vfinfo = NULL;
5526 		}
5527 		return err;
5528 	}
5529 
5530 	adap->num_vfs = num_vfs;
5531 	return num_vfs;
5532 }
5533 #endif /* CONFIG_PCI_IOV */
5534 
5535 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5536 {
5537 	struct net_device *netdev;
5538 	struct adapter *adapter;
5539 	static int adap_idx = 1;
5540 	int s_qpp, qpp, num_seg;
5541 	struct port_info *pi;
5542 	bool highdma = false;
5543 	enum chip_type chip;
5544 	void __iomem *regs;
5545 	int func, chip_ver;
5546 	u16 device_id;
5547 	int i, err;
5548 	u32 whoami;
5549 
5550 	printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5551 
5552 	err = pci_request_regions(pdev, KBUILD_MODNAME);
5553 	if (err) {
5554 		/* Just info, some other driver may have claimed the device. */
5555 		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5556 		return err;
5557 	}
5558 
5559 	err = pci_enable_device(pdev);
5560 	if (err) {
5561 		dev_err(&pdev->dev, "cannot enable PCI device\n");
5562 		goto out_release_regions;
5563 	}
5564 
5565 	regs = pci_ioremap_bar(pdev, 0);
5566 	if (!regs) {
5567 		dev_err(&pdev->dev, "cannot map device registers\n");
5568 		err = -ENOMEM;
5569 		goto out_disable_device;
5570 	}
5571 
5572 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5573 	if (!adapter) {
5574 		err = -ENOMEM;
5575 		goto out_unmap_bar0;
5576 	}
5577 
5578 	adapter->regs = regs;
5579 	err = t4_wait_dev_ready(regs);
5580 	if (err < 0)
5581 		goto out_free_adapter;
5582 
5583 	/* We control everything through one PF */
5584 	whoami = t4_read_reg(adapter, PL_WHOAMI_A);
5585 	pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
5586 	chip = t4_get_chip_type(adapter, CHELSIO_PCI_ID_VER(device_id));
5587 	if (chip < 0) {
5588 		dev_err(&pdev->dev, "Device %d is not supported\n", device_id);
5589 		err = chip;
5590 		goto out_free_adapter;
5591 	}
5592 	chip_ver = CHELSIO_CHIP_VERSION(chip);
5593 	func = chip_ver <= CHELSIO_T5 ?
5594 	       SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
5595 
5596 	adapter->pdev = pdev;
5597 	adapter->pdev_dev = &pdev->dev;
5598 	adapter->name = pci_name(pdev);
5599 	adapter->mbox = func;
5600 	adapter->pf = func;
5601 	adapter->params.chip = chip;
5602 	adapter->adap_idx = adap_idx;
5603 	adapter->msg_enable = DFLT_MSG_ENABLE;
5604 	adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5605 				    (sizeof(struct mbox_cmd) *
5606 				     T4_OS_LOG_MBOX_CMDS),
5607 				    GFP_KERNEL);
5608 	if (!adapter->mbox_log) {
5609 		err = -ENOMEM;
5610 		goto out_free_adapter;
5611 	}
5612 	spin_lock_init(&adapter->mbox_lock);
5613 	INIT_LIST_HEAD(&adapter->mlist.list);
5614 	adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
5615 	pci_set_drvdata(pdev, adapter);
5616 
5617 	if (func != ent->driver_data) {
5618 		pci_disable_device(pdev);
5619 		pci_save_state(pdev);        /* to restore SR-IOV later */
5620 		return 0;
5621 	}
5622 
5623 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5624 		highdma = true;
5625 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5626 		if (err) {
5627 			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5628 				"coherent allocations\n");
5629 			goto out_free_adapter;
5630 		}
5631 	} else {
5632 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5633 		if (err) {
5634 			dev_err(&pdev->dev, "no usable DMA configuration\n");
5635 			goto out_free_adapter;
5636 		}
5637 	}
5638 
5639 	pci_enable_pcie_error_reporting(pdev);
5640 	pci_set_master(pdev);
5641 	pci_save_state(pdev);
5642 	adap_idx++;
5643 	adapter->workq = create_singlethread_workqueue("cxgb4");
5644 	if (!adapter->workq) {
5645 		err = -ENOMEM;
5646 		goto out_free_adapter;
5647 	}
5648 
5649 	/* PCI device has been enabled */
5650 	adapter->flags |= CXGB4_DEV_ENABLED;
5651 	memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5652 
5653 	/* If possible, we use PCIe Relaxed Ordering Attribute to deliver
5654 	 * Ingress Packet Data to Free List Buffers in order to allow for
5655 	 * chipset performance optimizations between the Root Complex and
5656 	 * Memory Controllers.  (Messages to the associated Ingress Queue
5657 	 * notifying new Packet Placement in the Free Lists Buffers will be
5658 	 * send without the Relaxed Ordering Attribute thus guaranteeing that
5659 	 * all preceding PCIe Transaction Layer Packets will be processed
5660 	 * first.)  But some Root Complexes have various issues with Upstream
5661 	 * Transaction Layer Packets with the Relaxed Ordering Attribute set.
5662 	 * The PCIe devices which under the Root Complexes will be cleared the
5663 	 * Relaxed Ordering bit in the configuration space, So we check our
5664 	 * PCIe configuration space to see if it's flagged with advice against
5665 	 * using Relaxed Ordering.
5666 	 */
5667 	if (!pcie_relaxed_ordering_enabled(pdev))
5668 		adapter->flags |= CXGB4_ROOT_NO_RELAXED_ORDERING;
5669 
5670 	spin_lock_init(&adapter->stats_lock);
5671 	spin_lock_init(&adapter->tid_release_lock);
5672 	spin_lock_init(&adapter->win0_lock);
5673 
5674 	INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5675 	INIT_WORK(&adapter->db_full_task, process_db_full);
5676 	INIT_WORK(&adapter->db_drop_task, process_db_drop);
5677 	INIT_WORK(&adapter->fatal_err_notify_task, notify_fatal_err);
5678 
5679 	err = t4_prep_adapter(adapter);
5680 	if (err)
5681 		goto out_free_adapter;
5682 
5683 	if (is_kdump_kernel()) {
5684 		/* Collect hardware state and append to /proc/vmcore */
5685 		err = cxgb4_cudbg_vmcore_add_dump(adapter);
5686 		if (err) {
5687 			dev_warn(adapter->pdev_dev,
5688 				 "Fail collecting vmcore device dump, err: %d. Continuing\n",
5689 				 err);
5690 			err = 0;
5691 		}
5692 	}
5693 
5694 	if (!is_t4(adapter->params.chip)) {
5695 		s_qpp = (QUEUESPERPAGEPF0_S +
5696 			(QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
5697 			adapter->pf);
5698 		qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
5699 		      SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
5700 		num_seg = PAGE_SIZE / SEGMENT_SIZE;
5701 
5702 		/* Each segment size is 128B. Write coalescing is enabled only
5703 		 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5704 		 * queue is less no of segments that can be accommodated in
5705 		 * a page size.
5706 		 */
5707 		if (qpp > num_seg) {
5708 			dev_err(&pdev->dev,
5709 				"Incorrect number of egress queues per page\n");
5710 			err = -EINVAL;
5711 			goto out_free_adapter;
5712 		}
5713 		adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5714 		pci_resource_len(pdev, 2));
5715 		if (!adapter->bar2) {
5716 			dev_err(&pdev->dev, "cannot map device bar2 region\n");
5717 			err = -ENOMEM;
5718 			goto out_free_adapter;
5719 		}
5720 	}
5721 
5722 	setup_memwin(adapter);
5723 	err = adap_init0(adapter);
5724 #ifdef CONFIG_DEBUG_FS
5725 	bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
5726 #endif
5727 	setup_memwin_rdma(adapter);
5728 	if (err)
5729 		goto out_unmap_bar;
5730 
5731 	/* configure SGE_STAT_CFG_A to read WC stats */
5732 	if (!is_t4(adapter->params.chip))
5733 		t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
5734 			     (is_t5(adapter->params.chip) ? STATMODE_V(0) :
5735 			      T6_STATMODE_V(0)));
5736 
5737 	/* Initialize hash mac addr list */
5738 	INIT_LIST_HEAD(&adapter->mac_hlist);
5739 
5740 	for_each_port(adapter, i) {
5741 		netdev = alloc_etherdev_mq(sizeof(struct port_info),
5742 					   MAX_ETH_QSETS);
5743 		if (!netdev) {
5744 			err = -ENOMEM;
5745 			goto out_free_dev;
5746 		}
5747 
5748 		SET_NETDEV_DEV(netdev, &pdev->dev);
5749 
5750 		adapter->port[i] = netdev;
5751 		pi = netdev_priv(netdev);
5752 		pi->adapter = adapter;
5753 		pi->xact_addr_filt = -1;
5754 		pi->port_id = i;
5755 		netdev->irq = pdev->irq;
5756 
5757 		netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5758 			NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5759 			NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_GRO |
5760 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
5761 			NETIF_F_HW_TC;
5762 
5763 		if (chip_ver > CHELSIO_T5) {
5764 			netdev->hw_enc_features |= NETIF_F_IP_CSUM |
5765 						   NETIF_F_IPV6_CSUM |
5766 						   NETIF_F_RXCSUM |
5767 						   NETIF_F_GSO_UDP_TUNNEL |
5768 						   NETIF_F_GSO_UDP_TUNNEL_CSUM |
5769 						   NETIF_F_TSO | NETIF_F_TSO6;
5770 
5771 			netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
5772 					       NETIF_F_GSO_UDP_TUNNEL_CSUM |
5773 					       NETIF_F_HW_TLS_RECORD;
5774 		}
5775 
5776 		if (highdma)
5777 			netdev->hw_features |= NETIF_F_HIGHDMA;
5778 		netdev->features |= netdev->hw_features;
5779 		netdev->vlan_features = netdev->features & VLAN_FEAT;
5780 
5781 		netdev->priv_flags |= IFF_UNICAST_FLT;
5782 
5783 		/* MTU range: 81 - 9600 */
5784 		netdev->min_mtu = 81;              /* accommodate SACK */
5785 		netdev->max_mtu = MAX_MTU;
5786 
5787 		netdev->netdev_ops = &cxgb4_netdev_ops;
5788 #ifdef CONFIG_CHELSIO_T4_DCB
5789 		netdev->dcbnl_ops = &cxgb4_dcb_ops;
5790 		cxgb4_dcb_state_init(netdev);
5791 		cxgb4_dcb_version_init(netdev);
5792 #endif
5793 		cxgb4_set_ethtool_ops(netdev);
5794 	}
5795 
5796 	cxgb4_init_ethtool_dump(adapter);
5797 
5798 	pci_set_drvdata(pdev, adapter);
5799 
5800 	if (adapter->flags & CXGB4_FW_OK) {
5801 		err = t4_port_init(adapter, func, func, 0);
5802 		if (err)
5803 			goto out_free_dev;
5804 	} else if (adapter->params.nports == 1) {
5805 		/* If we don't have a connection to the firmware -- possibly
5806 		 * because of an error -- grab the raw VPD parameters so we
5807 		 * can set the proper MAC Address on the debug network
5808 		 * interface that we've created.
5809 		 */
5810 		u8 hw_addr[ETH_ALEN];
5811 		u8 *na = adapter->params.vpd.na;
5812 
5813 		err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
5814 		if (!err) {
5815 			for (i = 0; i < ETH_ALEN; i++)
5816 				hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
5817 					      hex2val(na[2 * i + 1]));
5818 			t4_set_hw_addr(adapter, 0, hw_addr);
5819 		}
5820 	}
5821 
5822 	if (!(adapter->flags & CXGB4_FW_OK))
5823 		goto fw_attach_fail;
5824 
5825 	/* Configure queues and allocate tables now, they can be needed as
5826 	 * soon as the first register_netdev completes.
5827 	 */
5828 	err = cfg_queues(adapter);
5829 	if (err)
5830 		goto out_free_dev;
5831 
5832 	adapter->smt = t4_init_smt();
5833 	if (!adapter->smt) {
5834 		/* We tolerate a lack of SMT, giving up some functionality */
5835 		dev_warn(&pdev->dev, "could not allocate SMT, continuing\n");
5836 	}
5837 
5838 	adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
5839 	if (!adapter->l2t) {
5840 		/* We tolerate a lack of L2T, giving up some functionality */
5841 		dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5842 		adapter->params.offload = 0;
5843 	}
5844 
5845 	adapter->mps_encap = kvcalloc(adapter->params.arch.mps_tcam_size,
5846 				      sizeof(struct mps_encap_entry),
5847 				      GFP_KERNEL);
5848 	if (!adapter->mps_encap)
5849 		dev_warn(&pdev->dev, "could not allocate MPS Encap entries, continuing\n");
5850 
5851 #if IS_ENABLED(CONFIG_IPV6)
5852 	if (chip_ver <= CHELSIO_T5 &&
5853 	    (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
5854 		/* CLIP functionality is not present in hardware,
5855 		 * hence disable all offload features
5856 		 */
5857 		dev_warn(&pdev->dev,
5858 			 "CLIP not enabled in hardware, continuing\n");
5859 		adapter->params.offload = 0;
5860 	} else {
5861 		adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
5862 						  adapter->clipt_end);
5863 		if (!adapter->clipt) {
5864 			/* We tolerate a lack of clip_table, giving up
5865 			 * some functionality
5866 			 */
5867 			dev_warn(&pdev->dev,
5868 				 "could not allocate Clip table, continuing\n");
5869 			adapter->params.offload = 0;
5870 		}
5871 	}
5872 #endif
5873 
5874 	for_each_port(adapter, i) {
5875 		pi = adap2pinfo(adapter, i);
5876 		pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
5877 		if (!pi->sched_tbl)
5878 			dev_warn(&pdev->dev,
5879 				 "could not activate scheduling on port %d\n",
5880 				 i);
5881 	}
5882 
5883 	if (tid_init(&adapter->tids) < 0) {
5884 		dev_warn(&pdev->dev, "could not allocate TID table, "
5885 			 "continuing\n");
5886 		adapter->params.offload = 0;
5887 	} else {
5888 		adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
5889 		if (!adapter->tc_u32)
5890 			dev_warn(&pdev->dev,
5891 				 "could not offload tc u32, continuing\n");
5892 
5893 		if (cxgb4_init_tc_flower(adapter))
5894 			dev_warn(&pdev->dev,
5895 				 "could not offload tc flower, continuing\n");
5896 	}
5897 
5898 	if (is_offload(adapter) || is_hashfilter(adapter)) {
5899 		if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
5900 			u32 hash_base, hash_reg;
5901 
5902 			if (chip_ver <= CHELSIO_T5) {
5903 				hash_reg = LE_DB_TID_HASHBASE_A;
5904 				hash_base = t4_read_reg(adapter, hash_reg);
5905 				adapter->tids.hash_base = hash_base / 4;
5906 			} else {
5907 				hash_reg = T6_LE_DB_HASH_TID_BASE_A;
5908 				hash_base = t4_read_reg(adapter, hash_reg);
5909 				adapter->tids.hash_base = hash_base;
5910 			}
5911 		}
5912 	}
5913 
5914 	/* See what interrupts we'll be using */
5915 	if (msi > 1 && enable_msix(adapter) == 0)
5916 		adapter->flags |= CXGB4_USING_MSIX;
5917 	else if (msi > 0 && pci_enable_msi(pdev) == 0) {
5918 		adapter->flags |= CXGB4_USING_MSI;
5919 		if (msi > 1)
5920 			free_msix_info(adapter);
5921 	}
5922 
5923 	/* check for PCI Express bandwidth capabiltites */
5924 	pcie_print_link_status(pdev);
5925 
5926 	err = init_rss(adapter);
5927 	if (err)
5928 		goto out_free_dev;
5929 
5930 	err = setup_fw_sge_queues(adapter);
5931 	if (err) {
5932 		dev_err(adapter->pdev_dev,
5933 			"FW sge queue allocation failed, err %d", err);
5934 		goto out_free_dev;
5935 	}
5936 
5937 fw_attach_fail:
5938 	/*
5939 	 * The card is now ready to go.  If any errors occur during device
5940 	 * registration we do not fail the whole card but rather proceed only
5941 	 * with the ports we manage to register successfully.  However we must
5942 	 * register at least one net device.
5943 	 */
5944 	for_each_port(adapter, i) {
5945 		pi = adap2pinfo(adapter, i);
5946 		adapter->port[i]->dev_port = pi->lport;
5947 		netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5948 		netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5949 
5950 		netif_carrier_off(adapter->port[i]);
5951 
5952 		err = register_netdev(adapter->port[i]);
5953 		if (err)
5954 			break;
5955 		adapter->chan_map[pi->tx_chan] = i;
5956 		print_port_info(adapter->port[i]);
5957 	}
5958 	if (i == 0) {
5959 		dev_err(&pdev->dev, "could not register any net devices\n");
5960 		goto out_free_dev;
5961 	}
5962 	if (err) {
5963 		dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5964 		err = 0;
5965 	}
5966 
5967 	if (cxgb4_debugfs_root) {
5968 		adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5969 							   cxgb4_debugfs_root);
5970 		setup_debugfs(adapter);
5971 	}
5972 
5973 	/* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5974 	pdev->needs_freset = 1;
5975 
5976 	if (is_uld(adapter)) {
5977 		mutex_lock(&uld_mutex);
5978 		list_add_tail(&adapter->list_node, &adapter_list);
5979 		mutex_unlock(&uld_mutex);
5980 	}
5981 
5982 	if (!is_t4(adapter->params.chip))
5983 		cxgb4_ptp_init(adapter);
5984 
5985 	if (IS_REACHABLE(CONFIG_THERMAL) &&
5986 	    !is_t4(adapter->params.chip) && (adapter->flags & CXGB4_FW_OK))
5987 		cxgb4_thermal_init(adapter);
5988 
5989 	print_adapter_info(adapter);
5990 	return 0;
5991 
5992  out_free_dev:
5993 	t4_free_sge_resources(adapter);
5994 	free_some_resources(adapter);
5995 	if (adapter->flags & CXGB4_USING_MSIX)
5996 		free_msix_info(adapter);
5997 	if (adapter->num_uld || adapter->num_ofld_uld)
5998 		t4_uld_mem_free(adapter);
5999  out_unmap_bar:
6000 	if (!is_t4(adapter->params.chip))
6001 		iounmap(adapter->bar2);
6002  out_free_adapter:
6003 	if (adapter->workq)
6004 		destroy_workqueue(adapter->workq);
6005 
6006 	kfree(adapter->mbox_log);
6007 	kfree(adapter);
6008  out_unmap_bar0:
6009 	iounmap(regs);
6010  out_disable_device:
6011 	pci_disable_pcie_error_reporting(pdev);
6012 	pci_disable_device(pdev);
6013  out_release_regions:
6014 	pci_release_regions(pdev);
6015 	return err;
6016 }
6017 
6018 static void remove_one(struct pci_dev *pdev)
6019 {
6020 	struct adapter *adapter = pci_get_drvdata(pdev);
6021 	struct hash_mac_addr *entry, *tmp;
6022 
6023 	if (!adapter) {
6024 		pci_release_regions(pdev);
6025 		return;
6026 	}
6027 
6028 	adapter->flags |= CXGB4_SHUTTING_DOWN;
6029 
6030 	if (adapter->pf == 4) {
6031 		int i;
6032 
6033 		/* Tear down per-adapter Work Queue first since it can contain
6034 		 * references to our adapter data structure.
6035 		 */
6036 		destroy_workqueue(adapter->workq);
6037 
6038 		if (is_uld(adapter)) {
6039 			detach_ulds(adapter);
6040 			t4_uld_clean_up(adapter);
6041 		}
6042 
6043 		adap_free_hma_mem(adapter);
6044 
6045 		disable_interrupts(adapter);
6046 
6047 		for_each_port(adapter, i)
6048 			if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6049 				unregister_netdev(adapter->port[i]);
6050 
6051 		debugfs_remove_recursive(adapter->debugfs_root);
6052 
6053 		if (!is_t4(adapter->params.chip))
6054 			cxgb4_ptp_stop(adapter);
6055 		if (IS_REACHABLE(CONFIG_THERMAL))
6056 			cxgb4_thermal_remove(adapter);
6057 
6058 		/* If we allocated filters, free up state associated with any
6059 		 * valid filters ...
6060 		 */
6061 		clear_all_filters(adapter);
6062 
6063 		if (adapter->flags & CXGB4_FULL_INIT_DONE)
6064 			cxgb_down(adapter);
6065 
6066 		if (adapter->flags & CXGB4_USING_MSIX)
6067 			free_msix_info(adapter);
6068 		if (adapter->num_uld || adapter->num_ofld_uld)
6069 			t4_uld_mem_free(adapter);
6070 		free_some_resources(adapter);
6071 		list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist,
6072 					 list) {
6073 			list_del(&entry->list);
6074 			kfree(entry);
6075 		}
6076 
6077 #if IS_ENABLED(CONFIG_IPV6)
6078 		t4_cleanup_clip_tbl(adapter);
6079 #endif
6080 		if (!is_t4(adapter->params.chip))
6081 			iounmap(adapter->bar2);
6082 	}
6083 #ifdef CONFIG_PCI_IOV
6084 	else {
6085 		cxgb4_iov_configure(adapter->pdev, 0);
6086 	}
6087 #endif
6088 	iounmap(adapter->regs);
6089 	pci_disable_pcie_error_reporting(pdev);
6090 	if ((adapter->flags & CXGB4_DEV_ENABLED)) {
6091 		pci_disable_device(pdev);
6092 		adapter->flags &= ~CXGB4_DEV_ENABLED;
6093 	}
6094 	pci_release_regions(pdev);
6095 	kfree(adapter->mbox_log);
6096 	synchronize_rcu();
6097 	kfree(adapter);
6098 }
6099 
6100 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
6101  * delivery.  This is essentially a stripped down version of the PCI remove()
6102  * function where we do the minimal amount of work necessary to shutdown any
6103  * further activity.
6104  */
6105 static void shutdown_one(struct pci_dev *pdev)
6106 {
6107 	struct adapter *adapter = pci_get_drvdata(pdev);
6108 
6109 	/* As with remove_one() above (see extended comment), we only want do
6110 	 * do cleanup on PCI Devices which went all the way through init_one()
6111 	 * ...
6112 	 */
6113 	if (!adapter) {
6114 		pci_release_regions(pdev);
6115 		return;
6116 	}
6117 
6118 	adapter->flags |= CXGB4_SHUTTING_DOWN;
6119 
6120 	if (adapter->pf == 4) {
6121 		int i;
6122 
6123 		for_each_port(adapter, i)
6124 			if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6125 				cxgb_close(adapter->port[i]);
6126 
6127 		if (is_uld(adapter)) {
6128 			detach_ulds(adapter);
6129 			t4_uld_clean_up(adapter);
6130 		}
6131 
6132 		disable_interrupts(adapter);
6133 		disable_msi(adapter);
6134 
6135 		t4_sge_stop(adapter);
6136 		if (adapter->flags & CXGB4_FW_OK)
6137 			t4_fw_bye(adapter, adapter->mbox);
6138 	}
6139 }
6140 
6141 static struct pci_driver cxgb4_driver = {
6142 	.name     = KBUILD_MODNAME,
6143 	.id_table = cxgb4_pci_tbl,
6144 	.probe    = init_one,
6145 	.remove   = remove_one,
6146 	.shutdown = shutdown_one,
6147 #ifdef CONFIG_PCI_IOV
6148 	.sriov_configure = cxgb4_iov_configure,
6149 #endif
6150 	.err_handler = &cxgb4_eeh,
6151 };
6152 
6153 static int __init cxgb4_init_module(void)
6154 {
6155 	int ret;
6156 
6157 	/* Debugfs support is optional, just warn if this fails */
6158 	cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6159 	if (!cxgb4_debugfs_root)
6160 		pr_warn("could not create debugfs entry, continuing\n");
6161 
6162 	ret = pci_register_driver(&cxgb4_driver);
6163 	if (ret < 0)
6164 		debugfs_remove(cxgb4_debugfs_root);
6165 
6166 #if IS_ENABLED(CONFIG_IPV6)
6167 	if (!inet6addr_registered) {
6168 		register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6169 		inet6addr_registered = true;
6170 	}
6171 #endif
6172 
6173 	return ret;
6174 }
6175 
6176 static void __exit cxgb4_cleanup_module(void)
6177 {
6178 #if IS_ENABLED(CONFIG_IPV6)
6179 	if (inet6addr_registered) {
6180 		unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6181 		inet6addr_registered = false;
6182 	}
6183 #endif
6184 	pci_unregister_driver(&cxgb4_driver);
6185 	debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
6186 }
6187 
6188 module_init(cxgb4_init_module);
6189 module_exit(cxgb4_cleanup_module);
6190