xref: /linux/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c (revision 3e44c471a2dab210f7e9b1e5f7d4d54d52df59eb)
1 /*
2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
3  *
4  * Copyright (c) 2003-2014 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 <net/addrconf.h>
66 #include <asm/uaccess.h>
67 
68 #include "cxgb4.h"
69 #include "t4_regs.h"
70 #include "t4_values.h"
71 #include "t4_msg.h"
72 #include "t4fw_api.h"
73 #include "t4fw_version.h"
74 #include "cxgb4_dcb.h"
75 #include "cxgb4_debugfs.h"
76 #include "clip_tbl.h"
77 #include "l2t.h"
78 
79 char cxgb4_driver_name[] = KBUILD_MODNAME;
80 
81 #ifdef DRV_VERSION
82 #undef DRV_VERSION
83 #endif
84 #define DRV_VERSION "2.0.0-ko"
85 const char cxgb4_driver_version[] = DRV_VERSION;
86 #define DRV_DESC "Chelsio T4/T5 Network Driver"
87 
88 /* Host shadow copy of ingress filter entry.  This is in host native format
89  * and doesn't match the ordering or bit order, etc. of the hardware of the
90  * firmware command.  The use of bit-field structure elements is purely to
91  * remind ourselves of the field size limitations and save memory in the case
92  * where the filter table is large.
93  */
94 struct filter_entry {
95 	/* Administrative fields for filter.
96 	 */
97 	u32 valid:1;            /* filter allocated and valid */
98 	u32 locked:1;           /* filter is administratively locked */
99 
100 	u32 pending:1;          /* filter action is pending firmware reply */
101 	u32 smtidx:8;           /* Source MAC Table index for smac */
102 	struct l2t_entry *l2t;  /* Layer Two Table entry for dmac */
103 
104 	/* The filter itself.  Most of this is a straight copy of information
105 	 * provided by the extended ioctl().  Some fields are translated to
106 	 * internal forms -- for instance the Ingress Queue ID passed in from
107 	 * the ioctl() is translated into the Absolute Ingress Queue ID.
108 	 */
109 	struct ch_filter_specification fs;
110 };
111 
112 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
113 			 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
114 			 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
115 
116 /* Macros needed to support the PCI Device ID Table ...
117  */
118 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
119 	static const struct pci_device_id cxgb4_pci_tbl[] = {
120 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
121 
122 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
123  * called for both.
124  */
125 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
126 
127 #define CH_PCI_ID_TABLE_ENTRY(devid) \
128 		{PCI_VDEVICE(CHELSIO, (devid)), 4}
129 
130 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
131 		{ 0, } \
132 	}
133 
134 #include "t4_pci_id_tbl.h"
135 
136 #define FW4_FNAME "cxgb4/t4fw.bin"
137 #define FW5_FNAME "cxgb4/t5fw.bin"
138 #define FW6_FNAME "cxgb4/t6fw.bin"
139 #define FW4_CFNAME "cxgb4/t4-config.txt"
140 #define FW5_CFNAME "cxgb4/t5-config.txt"
141 #define FW6_CFNAME "cxgb4/t6-config.txt"
142 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
143 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
144 #define PHY_AQ1202_DEVICEID 0x4409
145 #define PHY_BCM84834_DEVICEID 0x4486
146 
147 MODULE_DESCRIPTION(DRV_DESC);
148 MODULE_AUTHOR("Chelsio Communications");
149 MODULE_LICENSE("Dual BSD/GPL");
150 MODULE_VERSION(DRV_VERSION);
151 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
152 MODULE_FIRMWARE(FW4_FNAME);
153 MODULE_FIRMWARE(FW5_FNAME);
154 
155 /*
156  * Normally we're willing to become the firmware's Master PF but will be happy
157  * if another PF has already become the Master and initialized the adapter.
158  * Setting "force_init" will cause this driver to forcibly establish itself as
159  * the Master PF and initialize the adapter.
160  */
161 static uint force_init;
162 
163 module_param(force_init, uint, 0644);
164 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
165 
166 /*
167  * Normally if the firmware we connect to has Configuration File support, we
168  * use that and only fall back to the old Driver-based initialization if the
169  * Configuration File fails for some reason.  If force_old_init is set, then
170  * we'll always use the old Driver-based initialization sequence.
171  */
172 static uint force_old_init;
173 
174 module_param(force_old_init, uint, 0644);
175 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence, deprecated"
176 		 " parameter");
177 
178 static int dflt_msg_enable = DFLT_MSG_ENABLE;
179 
180 module_param(dflt_msg_enable, int, 0644);
181 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
182 
183 /*
184  * The driver uses the best interrupt scheme available on a platform in the
185  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
186  * of these schemes the driver may consider as follows:
187  *
188  * msi = 2: choose from among all three options
189  * msi = 1: only consider MSI and INTx interrupts
190  * msi = 0: force INTx interrupts
191  */
192 static int msi = 2;
193 
194 module_param(msi, int, 0644);
195 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
196 
197 /*
198  * Queue interrupt hold-off timer values.  Queues default to the first of these
199  * upon creation.
200  */
201 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
202 
203 module_param_array(intr_holdoff, uint, NULL, 0644);
204 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
205 		 "0..4 in microseconds, deprecated parameter");
206 
207 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
208 
209 module_param_array(intr_cnt, uint, NULL, 0644);
210 MODULE_PARM_DESC(intr_cnt,
211 		 "thresholds 1..3 for queue interrupt packet counters, "
212 		 "deprecated parameter");
213 
214 /*
215  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
216  * offset by 2 bytes in order to have the IP headers line up on 4-byte
217  * boundaries.  This is a requirement for many architectures which will throw
218  * a machine check fault if an attempt is made to access one of the 4-byte IP
219  * header fields on a non-4-byte boundary.  And it's a major performance issue
220  * even on some architectures which allow it like some implementations of the
221  * x86 ISA.  However, some architectures don't mind this and for some very
222  * edge-case performance sensitive applications (like forwarding large volumes
223  * of small packets), setting this DMA offset to 0 will decrease the number of
224  * PCI-E Bus transfers enough to measurably affect performance.
225  */
226 static int rx_dma_offset = 2;
227 
228 static bool vf_acls;
229 
230 #ifdef CONFIG_PCI_IOV
231 module_param(vf_acls, bool, 0644);
232 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement, "
233 		 "deprecated parameter");
234 
235 /* Configure the number of PCI-E Virtual Function which are to be instantiated
236  * on SR-IOV Capable Physical Functions.
237  */
238 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV];
239 
240 module_param_array(num_vf, uint, NULL, 0644);
241 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
242 #endif
243 
244 /* TX Queue select used to determine what algorithm to use for selecting TX
245  * queue. Select between the kernel provided function (select_queue=0) or user
246  * cxgb_select_queue function (select_queue=1)
247  *
248  * Default: select_queue=0
249  */
250 static int select_queue;
251 module_param(select_queue, int, 0644);
252 MODULE_PARM_DESC(select_queue,
253 		 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
254 
255 static unsigned int tp_vlan_pri_map = HW_TPL_FR_MT_PR_IV_P_FC;
256 
257 module_param(tp_vlan_pri_map, uint, 0644);
258 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration, "
259 		 "deprecated parameter");
260 
261 static struct dentry *cxgb4_debugfs_root;
262 
263 static LIST_HEAD(adapter_list);
264 static DEFINE_MUTEX(uld_mutex);
265 /* Adapter list to be accessed from atomic context */
266 static LIST_HEAD(adap_rcu_list);
267 static DEFINE_SPINLOCK(adap_rcu_lock);
268 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
269 static const char *uld_str[] = { "RDMA", "iSCSI" };
270 
271 static void link_report(struct net_device *dev)
272 {
273 	if (!netif_carrier_ok(dev))
274 		netdev_info(dev, "link down\n");
275 	else {
276 		static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
277 
278 		const char *s = "10Mbps";
279 		const struct port_info *p = netdev_priv(dev);
280 
281 		switch (p->link_cfg.speed) {
282 		case 10000:
283 			s = "10Gbps";
284 			break;
285 		case 1000:
286 			s = "1000Mbps";
287 			break;
288 		case 100:
289 			s = "100Mbps";
290 			break;
291 		case 40000:
292 			s = "40Gbps";
293 			break;
294 		}
295 
296 		netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
297 			    fc[p->link_cfg.fc]);
298 	}
299 }
300 
301 #ifdef CONFIG_CHELSIO_T4_DCB
302 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
303 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
304 {
305 	struct port_info *pi = netdev_priv(dev);
306 	struct adapter *adap = pi->adapter;
307 	struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
308 	int i;
309 
310 	/* We use a simple mapping of Port TX Queue Index to DCB
311 	 * Priority when we're enabling DCB.
312 	 */
313 	for (i = 0; i < pi->nqsets; i++, txq++) {
314 		u32 name, value;
315 		int err;
316 
317 		name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
318 			FW_PARAMS_PARAM_X_V(
319 				FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
320 			FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
321 		value = enable ? i : 0xffffffff;
322 
323 		/* Since we can be called while atomic (from "interrupt
324 		 * level") we need to issue the Set Parameters Commannd
325 		 * without sleeping (timeout < 0).
326 		 */
327 		err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
328 					    &name, &value,
329 					    -FW_CMD_MAX_TIMEOUT);
330 
331 		if (err)
332 			dev_err(adap->pdev_dev,
333 				"Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
334 				enable ? "set" : "unset", pi->port_id, i, -err);
335 		else
336 			txq->dcb_prio = value;
337 	}
338 }
339 #endif /* CONFIG_CHELSIO_T4_DCB */
340 
341 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
342 {
343 	struct net_device *dev = adapter->port[port_id];
344 
345 	/* Skip changes from disabled ports. */
346 	if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
347 		if (link_stat)
348 			netif_carrier_on(dev);
349 		else {
350 #ifdef CONFIG_CHELSIO_T4_DCB
351 			cxgb4_dcb_state_init(dev);
352 			dcb_tx_queue_prio_enable(dev, false);
353 #endif /* CONFIG_CHELSIO_T4_DCB */
354 			netif_carrier_off(dev);
355 		}
356 
357 		link_report(dev);
358 	}
359 }
360 
361 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
362 {
363 	static const char *mod_str[] = {
364 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
365 	};
366 
367 	const struct net_device *dev = adap->port[port_id];
368 	const struct port_info *pi = netdev_priv(dev);
369 
370 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
371 		netdev_info(dev, "port module unplugged\n");
372 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
373 		netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
374 }
375 
376 /*
377  * Configure the exact and hash address filters to handle a port's multicast
378  * and secondary unicast MAC addresses.
379  */
380 static int set_addr_filters(const struct net_device *dev, bool sleep)
381 {
382 	u64 mhash = 0;
383 	u64 uhash = 0;
384 	bool free = true;
385 	u16 filt_idx[7];
386 	const u8 *addr[7];
387 	int ret, naddr = 0;
388 	const struct netdev_hw_addr *ha;
389 	int uc_cnt = netdev_uc_count(dev);
390 	int mc_cnt = netdev_mc_count(dev);
391 	const struct port_info *pi = netdev_priv(dev);
392 	unsigned int mb = pi->adapter->pf;
393 
394 	/* first do the secondary unicast addresses */
395 	netdev_for_each_uc_addr(ha, dev) {
396 		addr[naddr++] = ha->addr;
397 		if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
398 			ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
399 					naddr, addr, filt_idx, &uhash, sleep);
400 			if (ret < 0)
401 				return ret;
402 
403 			free = false;
404 			naddr = 0;
405 		}
406 	}
407 
408 	/* next set up the multicast addresses */
409 	netdev_for_each_mc_addr(ha, dev) {
410 		addr[naddr++] = ha->addr;
411 		if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
412 			ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
413 					naddr, addr, filt_idx, &mhash, sleep);
414 			if (ret < 0)
415 				return ret;
416 
417 			free = false;
418 			naddr = 0;
419 		}
420 	}
421 
422 	return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
423 				uhash | mhash, sleep);
424 }
425 
426 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
427 module_param(dbfifo_int_thresh, int, 0644);
428 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
429 
430 /*
431  * usecs to sleep while draining the dbfifo
432  */
433 static int dbfifo_drain_delay = 1000;
434 module_param(dbfifo_drain_delay, int, 0644);
435 MODULE_PARM_DESC(dbfifo_drain_delay,
436 		 "usecs to sleep while draining the dbfifo");
437 
438 /*
439  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
440  * If @mtu is -1 it is left unchanged.
441  */
442 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
443 {
444 	int ret;
445 	struct port_info *pi = netdev_priv(dev);
446 
447 	ret = set_addr_filters(dev, sleep_ok);
448 	if (ret == 0)
449 		ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, mtu,
450 				    (dev->flags & IFF_PROMISC) ? 1 : 0,
451 				    (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
452 				    sleep_ok);
453 	return ret;
454 }
455 
456 /**
457  *	link_start - enable a port
458  *	@dev: the port to enable
459  *
460  *	Performs the MAC and PHY actions needed to enable a port.
461  */
462 static int link_start(struct net_device *dev)
463 {
464 	int ret;
465 	struct port_info *pi = netdev_priv(dev);
466 	unsigned int mb = pi->adapter->pf;
467 
468 	/*
469 	 * We do not set address filters and promiscuity here, the stack does
470 	 * that step explicitly.
471 	 */
472 	ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
473 			    !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
474 	if (ret == 0) {
475 		ret = t4_change_mac(pi->adapter, mb, pi->viid,
476 				    pi->xact_addr_filt, dev->dev_addr, true,
477 				    true);
478 		if (ret >= 0) {
479 			pi->xact_addr_filt = ret;
480 			ret = 0;
481 		}
482 	}
483 	if (ret == 0)
484 		ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
485 				    &pi->link_cfg);
486 	if (ret == 0) {
487 		local_bh_disable();
488 		ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
489 					  true, CXGB4_DCB_ENABLED);
490 		local_bh_enable();
491 	}
492 
493 	return ret;
494 }
495 
496 int cxgb4_dcb_enabled(const struct net_device *dev)
497 {
498 #ifdef CONFIG_CHELSIO_T4_DCB
499 	struct port_info *pi = netdev_priv(dev);
500 
501 	if (!pi->dcb.enabled)
502 		return 0;
503 
504 	return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
505 		(pi->dcb.state == CXGB4_DCB_STATE_HOST));
506 #else
507 	return 0;
508 #endif
509 }
510 EXPORT_SYMBOL(cxgb4_dcb_enabled);
511 
512 #ifdef CONFIG_CHELSIO_T4_DCB
513 /* Handle a Data Center Bridging update message from the firmware. */
514 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
515 {
516 	int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
517 	struct net_device *dev = adap->port[port];
518 	int old_dcb_enabled = cxgb4_dcb_enabled(dev);
519 	int new_dcb_enabled;
520 
521 	cxgb4_dcb_handle_fw_update(adap, pcmd);
522 	new_dcb_enabled = cxgb4_dcb_enabled(dev);
523 
524 	/* If the DCB has become enabled or disabled on the port then we're
525 	 * going to need to set up/tear down DCB Priority parameters for the
526 	 * TX Queues associated with the port.
527 	 */
528 	if (new_dcb_enabled != old_dcb_enabled)
529 		dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
530 }
531 #endif /* CONFIG_CHELSIO_T4_DCB */
532 
533 /* Clear a filter and release any of its resources that we own.  This also
534  * clears the filter's "pending" status.
535  */
536 static void clear_filter(struct adapter *adap, struct filter_entry *f)
537 {
538 	/* If the new or old filter have loopback rewriteing rules then we'll
539 	 * need to free any existing Layer Two Table (L2T) entries of the old
540 	 * filter rule.  The firmware will handle freeing up any Source MAC
541 	 * Table (SMT) entries used for rewriting Source MAC Addresses in
542 	 * loopback rules.
543 	 */
544 	if (f->l2t)
545 		cxgb4_l2t_release(f->l2t);
546 
547 	/* The zeroing of the filter rule below clears the filter valid,
548 	 * pending, locked flags, l2t pointer, etc. so it's all we need for
549 	 * this operation.
550 	 */
551 	memset(f, 0, sizeof(*f));
552 }
553 
554 /* Handle a filter write/deletion reply.
555  */
556 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
557 {
558 	unsigned int idx = GET_TID(rpl);
559 	unsigned int nidx = idx - adap->tids.ftid_base;
560 	unsigned int ret;
561 	struct filter_entry *f;
562 
563 	if (idx >= adap->tids.ftid_base && nidx <
564 	   (adap->tids.nftids + adap->tids.nsftids)) {
565 		idx = nidx;
566 		ret = TCB_COOKIE_G(rpl->cookie);
567 		f = &adap->tids.ftid_tab[idx];
568 
569 		if (ret == FW_FILTER_WR_FLT_DELETED) {
570 			/* Clear the filter when we get confirmation from the
571 			 * hardware that the filter has been deleted.
572 			 */
573 			clear_filter(adap, f);
574 		} else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
575 			dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
576 				idx);
577 			clear_filter(adap, f);
578 		} else if (ret == FW_FILTER_WR_FLT_ADDED) {
579 			f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
580 			f->pending = 0;  /* asynchronous setup completed */
581 			f->valid = 1;
582 		} else {
583 			/* Something went wrong.  Issue a warning about the
584 			 * problem and clear everything out.
585 			 */
586 			dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
587 				idx, ret);
588 			clear_filter(adap, f);
589 		}
590 	}
591 }
592 
593 /* Response queue handler for the FW event queue.
594  */
595 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
596 			  const struct pkt_gl *gl)
597 {
598 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
599 
600 	rsp++;                                          /* skip RSS header */
601 
602 	/* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
603 	 */
604 	if (unlikely(opcode == CPL_FW4_MSG &&
605 	   ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
606 		rsp++;
607 		opcode = ((const struct rss_header *)rsp)->opcode;
608 		rsp++;
609 		if (opcode != CPL_SGE_EGR_UPDATE) {
610 			dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
611 				, opcode);
612 			goto out;
613 		}
614 	}
615 
616 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
617 		const struct cpl_sge_egr_update *p = (void *)rsp;
618 		unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
619 		struct sge_txq *txq;
620 
621 		txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
622 		txq->restarts++;
623 		if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
624 			struct sge_eth_txq *eq;
625 
626 			eq = container_of(txq, struct sge_eth_txq, q);
627 			netif_tx_wake_queue(eq->txq);
628 		} else {
629 			struct sge_ofld_txq *oq;
630 
631 			oq = container_of(txq, struct sge_ofld_txq, q);
632 			tasklet_schedule(&oq->qresume_tsk);
633 		}
634 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
635 		const struct cpl_fw6_msg *p = (void *)rsp;
636 
637 #ifdef CONFIG_CHELSIO_T4_DCB
638 		const struct fw_port_cmd *pcmd = (const void *)p->data;
639 		unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
640 		unsigned int action =
641 			FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
642 
643 		if (cmd == FW_PORT_CMD &&
644 		    action == FW_PORT_ACTION_GET_PORT_INFO) {
645 			int port = FW_PORT_CMD_PORTID_G(
646 					be32_to_cpu(pcmd->op_to_portid));
647 			struct net_device *dev = q->adap->port[port];
648 			int state_input = ((pcmd->u.info.dcbxdis_pkd &
649 					    FW_PORT_CMD_DCBXDIS_F)
650 					   ? CXGB4_DCB_INPUT_FW_DISABLED
651 					   : CXGB4_DCB_INPUT_FW_ENABLED);
652 
653 			cxgb4_dcb_state_fsm(dev, state_input);
654 		}
655 
656 		if (cmd == FW_PORT_CMD &&
657 		    action == FW_PORT_ACTION_L2_DCB_CFG)
658 			dcb_rpl(q->adap, pcmd);
659 		else
660 #endif
661 			if (p->type == 0)
662 				t4_handle_fw_rpl(q->adap, p->data);
663 	} else if (opcode == CPL_L2T_WRITE_RPL) {
664 		const struct cpl_l2t_write_rpl *p = (void *)rsp;
665 
666 		do_l2t_write_rpl(q->adap, p);
667 	} else if (opcode == CPL_SET_TCB_RPL) {
668 		const struct cpl_set_tcb_rpl *p = (void *)rsp;
669 
670 		filter_rpl(q->adap, p);
671 	} else
672 		dev_err(q->adap->pdev_dev,
673 			"unexpected CPL %#x on FW event queue\n", opcode);
674 out:
675 	return 0;
676 }
677 
678 /**
679  *	uldrx_handler - response queue handler for ULD queues
680  *	@q: the response queue that received the packet
681  *	@rsp: the response queue descriptor holding the offload message
682  *	@gl: the gather list of packet fragments
683  *
684  *	Deliver an ingress offload packet to a ULD.  All processing is done by
685  *	the ULD, we just maintain statistics.
686  */
687 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
688 			 const struct pkt_gl *gl)
689 {
690 	struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
691 
692 	/* FW can send CPLs encapsulated in a CPL_FW4_MSG.
693 	 */
694 	if (((const struct rss_header *)rsp)->opcode == CPL_FW4_MSG &&
695 	    ((const struct cpl_fw4_msg *)(rsp + 1))->type == FW_TYPE_RSSCPL)
696 		rsp += 2;
697 
698 	if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
699 		rxq->stats.nomem++;
700 		return -1;
701 	}
702 	if (gl == NULL)
703 		rxq->stats.imm++;
704 	else if (gl == CXGB4_MSG_AN)
705 		rxq->stats.an++;
706 	else
707 		rxq->stats.pkts++;
708 	return 0;
709 }
710 
711 static void disable_msi(struct adapter *adapter)
712 {
713 	if (adapter->flags & USING_MSIX) {
714 		pci_disable_msix(adapter->pdev);
715 		adapter->flags &= ~USING_MSIX;
716 	} else if (adapter->flags & USING_MSI) {
717 		pci_disable_msi(adapter->pdev);
718 		adapter->flags &= ~USING_MSI;
719 	}
720 }
721 
722 /*
723  * Interrupt handler for non-data events used with MSI-X.
724  */
725 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
726 {
727 	struct adapter *adap = cookie;
728 	u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
729 
730 	if (v & PFSW_F) {
731 		adap->swintr = 1;
732 		t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
733 	}
734 	if (adap->flags & MASTER_PF)
735 		t4_slow_intr_handler(adap);
736 	return IRQ_HANDLED;
737 }
738 
739 /*
740  * Name the MSI-X interrupts.
741  */
742 static void name_msix_vecs(struct adapter *adap)
743 {
744 	int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
745 
746 	/* non-data interrupts */
747 	snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
748 
749 	/* FW events */
750 	snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
751 		 adap->port[0]->name);
752 
753 	/* Ethernet queues */
754 	for_each_port(adap, j) {
755 		struct net_device *d = adap->port[j];
756 		const struct port_info *pi = netdev_priv(d);
757 
758 		for (i = 0; i < pi->nqsets; i++, msi_idx++)
759 			snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
760 				 d->name, i);
761 	}
762 
763 	/* offload queues */
764 	for_each_ofldrxq(&adap->sge, i)
765 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
766 			 adap->port[0]->name, i);
767 
768 	for_each_rdmarxq(&adap->sge, i)
769 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
770 			 adap->port[0]->name, i);
771 
772 	for_each_rdmaciq(&adap->sge, i)
773 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma-ciq%d",
774 			 adap->port[0]->name, i);
775 }
776 
777 static int request_msix_queue_irqs(struct adapter *adap)
778 {
779 	struct sge *s = &adap->sge;
780 	int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, rdmaciqqidx = 0;
781 	int msi_index = 2;
782 
783 	err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
784 			  adap->msix_info[1].desc, &s->fw_evtq);
785 	if (err)
786 		return err;
787 
788 	for_each_ethrxq(s, ethqidx) {
789 		err = request_irq(adap->msix_info[msi_index].vec,
790 				  t4_sge_intr_msix, 0,
791 				  adap->msix_info[msi_index].desc,
792 				  &s->ethrxq[ethqidx].rspq);
793 		if (err)
794 			goto unwind;
795 		msi_index++;
796 	}
797 	for_each_ofldrxq(s, ofldqidx) {
798 		err = request_irq(adap->msix_info[msi_index].vec,
799 				  t4_sge_intr_msix, 0,
800 				  adap->msix_info[msi_index].desc,
801 				  &s->ofldrxq[ofldqidx].rspq);
802 		if (err)
803 			goto unwind;
804 		msi_index++;
805 	}
806 	for_each_rdmarxq(s, rdmaqidx) {
807 		err = request_irq(adap->msix_info[msi_index].vec,
808 				  t4_sge_intr_msix, 0,
809 				  adap->msix_info[msi_index].desc,
810 				  &s->rdmarxq[rdmaqidx].rspq);
811 		if (err)
812 			goto unwind;
813 		msi_index++;
814 	}
815 	for_each_rdmaciq(s, rdmaciqqidx) {
816 		err = request_irq(adap->msix_info[msi_index].vec,
817 				  t4_sge_intr_msix, 0,
818 				  adap->msix_info[msi_index].desc,
819 				  &s->rdmaciq[rdmaciqqidx].rspq);
820 		if (err)
821 			goto unwind;
822 		msi_index++;
823 	}
824 	return 0;
825 
826 unwind:
827 	while (--rdmaciqqidx >= 0)
828 		free_irq(adap->msix_info[--msi_index].vec,
829 			 &s->rdmaciq[rdmaciqqidx].rspq);
830 	while (--rdmaqidx >= 0)
831 		free_irq(adap->msix_info[--msi_index].vec,
832 			 &s->rdmarxq[rdmaqidx].rspq);
833 	while (--ofldqidx >= 0)
834 		free_irq(adap->msix_info[--msi_index].vec,
835 			 &s->ofldrxq[ofldqidx].rspq);
836 	while (--ethqidx >= 0)
837 		free_irq(adap->msix_info[--msi_index].vec,
838 			 &s->ethrxq[ethqidx].rspq);
839 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
840 	return err;
841 }
842 
843 static void free_msix_queue_irqs(struct adapter *adap)
844 {
845 	int i, msi_index = 2;
846 	struct sge *s = &adap->sge;
847 
848 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
849 	for_each_ethrxq(s, i)
850 		free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
851 	for_each_ofldrxq(s, i)
852 		free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
853 	for_each_rdmarxq(s, i)
854 		free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
855 	for_each_rdmaciq(s, i)
856 		free_irq(adap->msix_info[msi_index++].vec, &s->rdmaciq[i].rspq);
857 }
858 
859 /**
860  *	cxgb4_write_rss - write the RSS table for a given port
861  *	@pi: the port
862  *	@queues: array of queue indices for RSS
863  *
864  *	Sets up the portion of the HW RSS table for the port's VI to distribute
865  *	packets to the Rx queues in @queues.
866  *	Should never be called before setting up sge eth rx queues
867  */
868 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
869 {
870 	u16 *rss;
871 	int i, err;
872 	struct adapter *adapter = pi->adapter;
873 	const struct sge_eth_rxq *rxq;
874 
875 	rxq = &adapter->sge.ethrxq[pi->first_qset];
876 	rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
877 	if (!rss)
878 		return -ENOMEM;
879 
880 	/* map the queue indices to queue ids */
881 	for (i = 0; i < pi->rss_size; i++, queues++)
882 		rss[i] = rxq[*queues].rspq.abs_id;
883 
884 	err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
885 				  pi->rss_size, rss, pi->rss_size);
886 	/* If Tunnel All Lookup isn't specified in the global RSS
887 	 * Configuration, then we need to specify a default Ingress
888 	 * Queue for any ingress packets which aren't hashed.  We'll
889 	 * use our first ingress queue ...
890 	 */
891 	if (!err)
892 		err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
893 				       FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
894 				       FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
895 				       FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
896 				       FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
897 				       FW_RSS_VI_CONFIG_CMD_UDPEN_F,
898 				       rss[0]);
899 	kfree(rss);
900 	return err;
901 }
902 
903 /**
904  *	setup_rss - configure RSS
905  *	@adap: the adapter
906  *
907  *	Sets up RSS for each port.
908  */
909 static int setup_rss(struct adapter *adap)
910 {
911 	int i, j, err;
912 
913 	for_each_port(adap, i) {
914 		const struct port_info *pi = adap2pinfo(adap, i);
915 
916 		/* Fill default values with equal distribution */
917 		for (j = 0; j < pi->rss_size; j++)
918 			pi->rss[j] = j % pi->nqsets;
919 
920 		err = cxgb4_write_rss(pi, pi->rss);
921 		if (err)
922 			return err;
923 	}
924 	return 0;
925 }
926 
927 /*
928  * Return the channel of the ingress queue with the given qid.
929  */
930 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
931 {
932 	qid -= p->ingr_start;
933 	return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
934 }
935 
936 /*
937  * Wait until all NAPI handlers are descheduled.
938  */
939 static void quiesce_rx(struct adapter *adap)
940 {
941 	int i;
942 
943 	for (i = 0; i < adap->sge.ingr_sz; i++) {
944 		struct sge_rspq *q = adap->sge.ingr_map[i];
945 
946 		if (q && q->handler) {
947 			napi_disable(&q->napi);
948 			local_bh_disable();
949 			while (!cxgb_poll_lock_napi(q))
950 				mdelay(1);
951 			local_bh_enable();
952 		}
953 
954 	}
955 }
956 
957 /* Disable interrupt and napi handler */
958 static void disable_interrupts(struct adapter *adap)
959 {
960 	if (adap->flags & FULL_INIT_DONE) {
961 		t4_intr_disable(adap);
962 		if (adap->flags & USING_MSIX) {
963 			free_msix_queue_irqs(adap);
964 			free_irq(adap->msix_info[0].vec, adap);
965 		} else {
966 			free_irq(adap->pdev->irq, adap);
967 		}
968 		quiesce_rx(adap);
969 	}
970 }
971 
972 /*
973  * Enable NAPI scheduling and interrupt generation for all Rx queues.
974  */
975 static void enable_rx(struct adapter *adap)
976 {
977 	int i;
978 
979 	for (i = 0; i < adap->sge.ingr_sz; i++) {
980 		struct sge_rspq *q = adap->sge.ingr_map[i];
981 
982 		if (!q)
983 			continue;
984 		if (q->handler) {
985 			cxgb_busy_poll_init_lock(q);
986 			napi_enable(&q->napi);
987 		}
988 		/* 0-increment GTS to start the timer and enable interrupts */
989 		t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
990 			     SEINTARM_V(q->intr_params) |
991 			     INGRESSQID_V(q->cntxt_id));
992 	}
993 }
994 
995 static int alloc_ofld_rxqs(struct adapter *adap, struct sge_ofld_rxq *q,
996 			   unsigned int nq, unsigned int per_chan, int msi_idx,
997 			   u16 *ids)
998 {
999 	int i, err;
1000 
1001 	for (i = 0; i < nq; i++, q++) {
1002 		if (msi_idx > 0)
1003 			msi_idx++;
1004 		err = t4_sge_alloc_rxq(adap, &q->rspq, false,
1005 				       adap->port[i / per_chan],
1006 				       msi_idx, q->fl.size ? &q->fl : NULL,
1007 				       uldrx_handler, 0);
1008 		if (err)
1009 			return err;
1010 		memset(&q->stats, 0, sizeof(q->stats));
1011 		if (ids)
1012 			ids[i] = q->rspq.abs_id;
1013 	}
1014 	return 0;
1015 }
1016 
1017 /**
1018  *	setup_sge_queues - configure SGE Tx/Rx/response queues
1019  *	@adap: the adapter
1020  *
1021  *	Determines how many sets of SGE queues to use and initializes them.
1022  *	We support multiple queue sets per port if we have MSI-X, otherwise
1023  *	just one queue set per port.
1024  */
1025 static int setup_sge_queues(struct adapter *adap)
1026 {
1027 	int err, msi_idx, i, j;
1028 	struct sge *s = &adap->sge;
1029 
1030 	bitmap_zero(s->starving_fl, s->egr_sz);
1031 	bitmap_zero(s->txq_maperr, s->egr_sz);
1032 
1033 	if (adap->flags & USING_MSIX)
1034 		msi_idx = 1;         /* vector 0 is for non-queue interrupts */
1035 	else {
1036 		err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
1037 				       NULL, NULL, -1);
1038 		if (err)
1039 			return err;
1040 		msi_idx = -((int)s->intrq.abs_id + 1);
1041 	}
1042 
1043 	/* NOTE: If you add/delete any Ingress/Egress Queue allocations in here,
1044 	 * don't forget to update the following which need to be
1045 	 * synchronized to and changes here.
1046 	 *
1047 	 * 1. The calculations of MAX_INGQ in cxgb4.h.
1048 	 *
1049 	 * 2. Update enable_msix/name_msix_vecs/request_msix_queue_irqs
1050 	 *    to accommodate any new/deleted Ingress Queues
1051 	 *    which need MSI-X Vectors.
1052 	 *
1053 	 * 3. Update sge_qinfo_show() to include information on the
1054 	 *    new/deleted queues.
1055 	 */
1056 	err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
1057 			       msi_idx, NULL, fwevtq_handler, -1);
1058 	if (err) {
1059 freeout:	t4_free_sge_resources(adap);
1060 		return err;
1061 	}
1062 
1063 	for_each_port(adap, i) {
1064 		struct net_device *dev = adap->port[i];
1065 		struct port_info *pi = netdev_priv(dev);
1066 		struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1067 		struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1068 
1069 		for (j = 0; j < pi->nqsets; j++, q++) {
1070 			if (msi_idx > 0)
1071 				msi_idx++;
1072 			err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1073 					       msi_idx, &q->fl,
1074 					       t4_ethrx_handler,
1075 					       t4_get_mps_bg_map(adap,
1076 								 pi->tx_chan));
1077 			if (err)
1078 				goto freeout;
1079 			q->rspq.idx = j;
1080 			memset(&q->stats, 0, sizeof(q->stats));
1081 		}
1082 		for (j = 0; j < pi->nqsets; j++, t++) {
1083 			err = t4_sge_alloc_eth_txq(adap, t, dev,
1084 					netdev_get_tx_queue(dev, j),
1085 					s->fw_evtq.cntxt_id);
1086 			if (err)
1087 				goto freeout;
1088 		}
1089 	}
1090 
1091 	j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
1092 	for_each_ofldrxq(s, i) {
1093 		err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i],
1094 					    adap->port[i / j],
1095 					    s->fw_evtq.cntxt_id);
1096 		if (err)
1097 			goto freeout;
1098 	}
1099 
1100 #define ALLOC_OFLD_RXQS(firstq, nq, per_chan, ids) do { \
1101 	err = alloc_ofld_rxqs(adap, firstq, nq, per_chan, msi_idx, ids); \
1102 	if (err) \
1103 		goto freeout; \
1104 	if (msi_idx > 0) \
1105 		msi_idx += nq; \
1106 } while (0)
1107 
1108 	ALLOC_OFLD_RXQS(s->ofldrxq, s->ofldqsets, j, s->ofld_rxq);
1109 	ALLOC_OFLD_RXQS(s->rdmarxq, s->rdmaqs, 1, s->rdma_rxq);
1110 	j = s->rdmaciqs / adap->params.nports; /* rdmaq queues per channel */
1111 	ALLOC_OFLD_RXQS(s->rdmaciq, s->rdmaciqs, j, s->rdma_ciq);
1112 
1113 #undef ALLOC_OFLD_RXQS
1114 
1115 	for_each_port(adap, i) {
1116 		/*
1117 		 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1118 		 * have RDMA queues, and that's the right value.
1119 		 */
1120 		err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1121 					    s->fw_evtq.cntxt_id,
1122 					    s->rdmarxq[i].rspq.cntxt_id);
1123 		if (err)
1124 			goto freeout;
1125 	}
1126 
1127 	t4_write_reg(adap, is_t4(adap->params.chip) ?
1128 				MPS_TRC_RSS_CONTROL_A :
1129 				MPS_T5_TRC_RSS_CONTROL_A,
1130 		     RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
1131 		     QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
1132 	return 0;
1133 }
1134 
1135 /*
1136  * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1137  * The allocated memory is cleared.
1138  */
1139 void *t4_alloc_mem(size_t size)
1140 {
1141 	void *p = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1142 
1143 	if (!p)
1144 		p = vzalloc(size);
1145 	return p;
1146 }
1147 
1148 /*
1149  * Free memory allocated through alloc_mem().
1150  */
1151 void t4_free_mem(void *addr)
1152 {
1153 	if (is_vmalloc_addr(addr))
1154 		vfree(addr);
1155 	else
1156 		kfree(addr);
1157 }
1158 
1159 /* Send a Work Request to write the filter at a specified index.  We construct
1160  * a Firmware Filter Work Request to have the work done and put the indicated
1161  * filter into "pending" mode which will prevent any further actions against
1162  * it till we get a reply from the firmware on the completion status of the
1163  * request.
1164  */
1165 static int set_filter_wr(struct adapter *adapter, int fidx)
1166 {
1167 	struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1168 	struct sk_buff *skb;
1169 	struct fw_filter_wr *fwr;
1170 	unsigned int ftid;
1171 
1172 	skb = alloc_skb(sizeof(*fwr), GFP_KERNEL);
1173 	if (!skb)
1174 		return -ENOMEM;
1175 
1176 	/* If the new filter requires loopback Destination MAC and/or VLAN
1177 	 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1178 	 * the filter.
1179 	 */
1180 	if (f->fs.newdmac || f->fs.newvlan) {
1181 		/* allocate L2T entry for new filter */
1182 		f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1183 		if (f->l2t == NULL) {
1184 			kfree_skb(skb);
1185 			return -EAGAIN;
1186 		}
1187 		if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1188 					f->fs.eport, f->fs.dmac)) {
1189 			cxgb4_l2t_release(f->l2t);
1190 			f->l2t = NULL;
1191 			kfree_skb(skb);
1192 			return -ENOMEM;
1193 		}
1194 	}
1195 
1196 	ftid = adapter->tids.ftid_base + fidx;
1197 
1198 	fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1199 	memset(fwr, 0, sizeof(*fwr));
1200 
1201 	/* It would be nice to put most of the following in t4_hw.c but most
1202 	 * of the work is translating the cxgbtool ch_filter_specification
1203 	 * into the Work Request and the definition of that structure is
1204 	 * currently in cxgbtool.h which isn't appropriate to pull into the
1205 	 * common code.  We may eventually try to come up with a more neutral
1206 	 * filter specification structure but for now it's easiest to simply
1207 	 * put this fairly direct code in line ...
1208 	 */
1209 	fwr->op_pkd = htonl(FW_WR_OP_V(FW_FILTER_WR));
1210 	fwr->len16_pkd = htonl(FW_WR_LEN16_V(sizeof(*fwr)/16));
1211 	fwr->tid_to_iq =
1212 		htonl(FW_FILTER_WR_TID_V(ftid) |
1213 		      FW_FILTER_WR_RQTYPE_V(f->fs.type) |
1214 		      FW_FILTER_WR_NOREPLY_V(0) |
1215 		      FW_FILTER_WR_IQ_V(f->fs.iq));
1216 	fwr->del_filter_to_l2tix =
1217 		htonl(FW_FILTER_WR_RPTTID_V(f->fs.rpttid) |
1218 		      FW_FILTER_WR_DROP_V(f->fs.action == FILTER_DROP) |
1219 		      FW_FILTER_WR_DIRSTEER_V(f->fs.dirsteer) |
1220 		      FW_FILTER_WR_MASKHASH_V(f->fs.maskhash) |
1221 		      FW_FILTER_WR_DIRSTEERHASH_V(f->fs.dirsteerhash) |
1222 		      FW_FILTER_WR_LPBK_V(f->fs.action == FILTER_SWITCH) |
1223 		      FW_FILTER_WR_DMAC_V(f->fs.newdmac) |
1224 		      FW_FILTER_WR_SMAC_V(f->fs.newsmac) |
1225 		      FW_FILTER_WR_INSVLAN_V(f->fs.newvlan == VLAN_INSERT ||
1226 					     f->fs.newvlan == VLAN_REWRITE) |
1227 		      FW_FILTER_WR_RMVLAN_V(f->fs.newvlan == VLAN_REMOVE ||
1228 					    f->fs.newvlan == VLAN_REWRITE) |
1229 		      FW_FILTER_WR_HITCNTS_V(f->fs.hitcnts) |
1230 		      FW_FILTER_WR_TXCHAN_V(f->fs.eport) |
1231 		      FW_FILTER_WR_PRIO_V(f->fs.prio) |
1232 		      FW_FILTER_WR_L2TIX_V(f->l2t ? f->l2t->idx : 0));
1233 	fwr->ethtype = htons(f->fs.val.ethtype);
1234 	fwr->ethtypem = htons(f->fs.mask.ethtype);
1235 	fwr->frag_to_ovlan_vldm =
1236 		(FW_FILTER_WR_FRAG_V(f->fs.val.frag) |
1237 		 FW_FILTER_WR_FRAGM_V(f->fs.mask.frag) |
1238 		 FW_FILTER_WR_IVLAN_VLD_V(f->fs.val.ivlan_vld) |
1239 		 FW_FILTER_WR_OVLAN_VLD_V(f->fs.val.ovlan_vld) |
1240 		 FW_FILTER_WR_IVLAN_VLDM_V(f->fs.mask.ivlan_vld) |
1241 		 FW_FILTER_WR_OVLAN_VLDM_V(f->fs.mask.ovlan_vld));
1242 	fwr->smac_sel = 0;
1243 	fwr->rx_chan_rx_rpl_iq =
1244 		htons(FW_FILTER_WR_RX_CHAN_V(0) |
1245 		      FW_FILTER_WR_RX_RPL_IQ_V(adapter->sge.fw_evtq.abs_id));
1246 	fwr->maci_to_matchtypem =
1247 		htonl(FW_FILTER_WR_MACI_V(f->fs.val.macidx) |
1248 		      FW_FILTER_WR_MACIM_V(f->fs.mask.macidx) |
1249 		      FW_FILTER_WR_FCOE_V(f->fs.val.fcoe) |
1250 		      FW_FILTER_WR_FCOEM_V(f->fs.mask.fcoe) |
1251 		      FW_FILTER_WR_PORT_V(f->fs.val.iport) |
1252 		      FW_FILTER_WR_PORTM_V(f->fs.mask.iport) |
1253 		      FW_FILTER_WR_MATCHTYPE_V(f->fs.val.matchtype) |
1254 		      FW_FILTER_WR_MATCHTYPEM_V(f->fs.mask.matchtype));
1255 	fwr->ptcl = f->fs.val.proto;
1256 	fwr->ptclm = f->fs.mask.proto;
1257 	fwr->ttyp = f->fs.val.tos;
1258 	fwr->ttypm = f->fs.mask.tos;
1259 	fwr->ivlan = htons(f->fs.val.ivlan);
1260 	fwr->ivlanm = htons(f->fs.mask.ivlan);
1261 	fwr->ovlan = htons(f->fs.val.ovlan);
1262 	fwr->ovlanm = htons(f->fs.mask.ovlan);
1263 	memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1264 	memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1265 	memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1266 	memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1267 	fwr->lp = htons(f->fs.val.lport);
1268 	fwr->lpm = htons(f->fs.mask.lport);
1269 	fwr->fp = htons(f->fs.val.fport);
1270 	fwr->fpm = htons(f->fs.mask.fport);
1271 	if (f->fs.newsmac)
1272 		memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1273 
1274 	/* Mark the filter as "pending" and ship off the Filter Work Request.
1275 	 * When we get the Work Request Reply we'll clear the pending status.
1276 	 */
1277 	f->pending = 1;
1278 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1279 	t4_ofld_send(adapter, skb);
1280 	return 0;
1281 }
1282 
1283 /* Delete the filter at a specified index.
1284  */
1285 static int del_filter_wr(struct adapter *adapter, int fidx)
1286 {
1287 	struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1288 	struct sk_buff *skb;
1289 	struct fw_filter_wr *fwr;
1290 	unsigned int len, ftid;
1291 
1292 	len = sizeof(*fwr);
1293 	ftid = adapter->tids.ftid_base + fidx;
1294 
1295 	skb = alloc_skb(len, GFP_KERNEL);
1296 	if (!skb)
1297 		return -ENOMEM;
1298 
1299 	fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1300 	t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1301 
1302 	/* Mark the filter as "pending" and ship off the Filter Work Request.
1303 	 * When we get the Work Request Reply we'll clear the pending status.
1304 	 */
1305 	f->pending = 1;
1306 	t4_mgmt_tx(adapter, skb);
1307 	return 0;
1308 }
1309 
1310 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
1311 			     void *accel_priv, select_queue_fallback_t fallback)
1312 {
1313 	int txq;
1314 
1315 #ifdef CONFIG_CHELSIO_T4_DCB
1316 	/* If a Data Center Bridging has been successfully negotiated on this
1317 	 * link then we'll use the skb's priority to map it to a TX Queue.
1318 	 * The skb's priority is determined via the VLAN Tag Priority Code
1319 	 * Point field.
1320 	 */
1321 	if (cxgb4_dcb_enabled(dev)) {
1322 		u16 vlan_tci;
1323 		int err;
1324 
1325 		err = vlan_get_tag(skb, &vlan_tci);
1326 		if (unlikely(err)) {
1327 			if (net_ratelimit())
1328 				netdev_warn(dev,
1329 					    "TX Packet without VLAN Tag on DCB Link\n");
1330 			txq = 0;
1331 		} else {
1332 			txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1333 #ifdef CONFIG_CHELSIO_T4_FCOE
1334 			if (skb->protocol == htons(ETH_P_FCOE))
1335 				txq = skb->priority & 0x7;
1336 #endif /* CONFIG_CHELSIO_T4_FCOE */
1337 		}
1338 		return txq;
1339 	}
1340 #endif /* CONFIG_CHELSIO_T4_DCB */
1341 
1342 	if (select_queue) {
1343 		txq = (skb_rx_queue_recorded(skb)
1344 			? skb_get_rx_queue(skb)
1345 			: smp_processor_id());
1346 
1347 		while (unlikely(txq >= dev->real_num_tx_queues))
1348 			txq -= dev->real_num_tx_queues;
1349 
1350 		return txq;
1351 	}
1352 
1353 	return fallback(dev, skb) % dev->real_num_tx_queues;
1354 }
1355 
1356 static int closest_timer(const struct sge *s, int time)
1357 {
1358 	int i, delta, match = 0, min_delta = INT_MAX;
1359 
1360 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1361 		delta = time - s->timer_val[i];
1362 		if (delta < 0)
1363 			delta = -delta;
1364 		if (delta < min_delta) {
1365 			min_delta = delta;
1366 			match = i;
1367 		}
1368 	}
1369 	return match;
1370 }
1371 
1372 static int closest_thres(const struct sge *s, int thres)
1373 {
1374 	int i, delta, match = 0, min_delta = INT_MAX;
1375 
1376 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1377 		delta = thres - s->counter_val[i];
1378 		if (delta < 0)
1379 			delta = -delta;
1380 		if (delta < min_delta) {
1381 			min_delta = delta;
1382 			match = i;
1383 		}
1384 	}
1385 	return match;
1386 }
1387 
1388 /**
1389  *	cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1390  *	@q: the Rx queue
1391  *	@us: the hold-off time in us, or 0 to disable timer
1392  *	@cnt: the hold-off packet count, or 0 to disable counter
1393  *
1394  *	Sets an Rx queue's interrupt hold-off time and packet count.  At least
1395  *	one of the two needs to be enabled for the queue to generate interrupts.
1396  */
1397 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1398 			       unsigned int us, unsigned int cnt)
1399 {
1400 	struct adapter *adap = q->adap;
1401 
1402 	if ((us | cnt) == 0)
1403 		cnt = 1;
1404 
1405 	if (cnt) {
1406 		int err;
1407 		u32 v, new_idx;
1408 
1409 		new_idx = closest_thres(&adap->sge, cnt);
1410 		if (q->desc && q->pktcnt_idx != new_idx) {
1411 			/* the queue has already been created, update it */
1412 			v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1413 			    FW_PARAMS_PARAM_X_V(
1414 					FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1415 			    FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1416 			err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1417 					    &v, &new_idx);
1418 			if (err)
1419 				return err;
1420 		}
1421 		q->pktcnt_idx = new_idx;
1422 	}
1423 
1424 	us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1425 	q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1426 	return 0;
1427 }
1428 
1429 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1430 {
1431 	const struct port_info *pi = netdev_priv(dev);
1432 	netdev_features_t changed = dev->features ^ features;
1433 	int err;
1434 
1435 	if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1436 		return 0;
1437 
1438 	err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1439 			    -1, -1, -1,
1440 			    !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1441 	if (unlikely(err))
1442 		dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1443 	return err;
1444 }
1445 
1446 static int setup_debugfs(struct adapter *adap)
1447 {
1448 	if (IS_ERR_OR_NULL(adap->debugfs_root))
1449 		return -1;
1450 
1451 #ifdef CONFIG_DEBUG_FS
1452 	t4_setup_debugfs(adap);
1453 #endif
1454 	return 0;
1455 }
1456 
1457 /*
1458  * upper-layer driver support
1459  */
1460 
1461 /*
1462  * Allocate an active-open TID and set it to the supplied value.
1463  */
1464 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1465 {
1466 	int atid = -1;
1467 
1468 	spin_lock_bh(&t->atid_lock);
1469 	if (t->afree) {
1470 		union aopen_entry *p = t->afree;
1471 
1472 		atid = (p - t->atid_tab) + t->atid_base;
1473 		t->afree = p->next;
1474 		p->data = data;
1475 		t->atids_in_use++;
1476 	}
1477 	spin_unlock_bh(&t->atid_lock);
1478 	return atid;
1479 }
1480 EXPORT_SYMBOL(cxgb4_alloc_atid);
1481 
1482 /*
1483  * Release an active-open TID.
1484  */
1485 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1486 {
1487 	union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1488 
1489 	spin_lock_bh(&t->atid_lock);
1490 	p->next = t->afree;
1491 	t->afree = p;
1492 	t->atids_in_use--;
1493 	spin_unlock_bh(&t->atid_lock);
1494 }
1495 EXPORT_SYMBOL(cxgb4_free_atid);
1496 
1497 /*
1498  * Allocate a server TID and set it to the supplied value.
1499  */
1500 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1501 {
1502 	int stid;
1503 
1504 	spin_lock_bh(&t->stid_lock);
1505 	if (family == PF_INET) {
1506 		stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1507 		if (stid < t->nstids)
1508 			__set_bit(stid, t->stid_bmap);
1509 		else
1510 			stid = -1;
1511 	} else {
1512 		stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
1513 		if (stid < 0)
1514 			stid = -1;
1515 	}
1516 	if (stid >= 0) {
1517 		t->stid_tab[stid].data = data;
1518 		stid += t->stid_base;
1519 		/* IPv6 requires max of 520 bits or 16 cells in TCAM
1520 		 * This is equivalent to 4 TIDs. With CLIP enabled it
1521 		 * needs 2 TIDs.
1522 		 */
1523 		if (family == PF_INET)
1524 			t->stids_in_use++;
1525 		else
1526 			t->stids_in_use += 4;
1527 	}
1528 	spin_unlock_bh(&t->stid_lock);
1529 	return stid;
1530 }
1531 EXPORT_SYMBOL(cxgb4_alloc_stid);
1532 
1533 /* Allocate a server filter TID and set it to the supplied value.
1534  */
1535 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1536 {
1537 	int stid;
1538 
1539 	spin_lock_bh(&t->stid_lock);
1540 	if (family == PF_INET) {
1541 		stid = find_next_zero_bit(t->stid_bmap,
1542 				t->nstids + t->nsftids, t->nstids);
1543 		if (stid < (t->nstids + t->nsftids))
1544 			__set_bit(stid, t->stid_bmap);
1545 		else
1546 			stid = -1;
1547 	} else {
1548 		stid = -1;
1549 	}
1550 	if (stid >= 0) {
1551 		t->stid_tab[stid].data = data;
1552 		stid -= t->nstids;
1553 		stid += t->sftid_base;
1554 		t->stids_in_use++;
1555 	}
1556 	spin_unlock_bh(&t->stid_lock);
1557 	return stid;
1558 }
1559 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1560 
1561 /* Release a server TID.
1562  */
1563 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1564 {
1565 	/* Is it a server filter TID? */
1566 	if (t->nsftids && (stid >= t->sftid_base)) {
1567 		stid -= t->sftid_base;
1568 		stid += t->nstids;
1569 	} else {
1570 		stid -= t->stid_base;
1571 	}
1572 
1573 	spin_lock_bh(&t->stid_lock);
1574 	if (family == PF_INET)
1575 		__clear_bit(stid, t->stid_bmap);
1576 	else
1577 		bitmap_release_region(t->stid_bmap, stid, 2);
1578 	t->stid_tab[stid].data = NULL;
1579 	if (family == PF_INET)
1580 		t->stids_in_use--;
1581 	else
1582 		t->stids_in_use -= 4;
1583 	spin_unlock_bh(&t->stid_lock);
1584 }
1585 EXPORT_SYMBOL(cxgb4_free_stid);
1586 
1587 /*
1588  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
1589  */
1590 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1591 			   unsigned int tid)
1592 {
1593 	struct cpl_tid_release *req;
1594 
1595 	set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1596 	req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
1597 	INIT_TP_WR(req, tid);
1598 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1599 }
1600 
1601 /*
1602  * Queue a TID release request and if necessary schedule a work queue to
1603  * process it.
1604  */
1605 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1606 				    unsigned int tid)
1607 {
1608 	void **p = &t->tid_tab[tid];
1609 	struct adapter *adap = container_of(t, struct adapter, tids);
1610 
1611 	spin_lock_bh(&adap->tid_release_lock);
1612 	*p = adap->tid_release_head;
1613 	/* Low 2 bits encode the Tx channel number */
1614 	adap->tid_release_head = (void **)((uintptr_t)p | chan);
1615 	if (!adap->tid_release_task_busy) {
1616 		adap->tid_release_task_busy = true;
1617 		queue_work(adap->workq, &adap->tid_release_task);
1618 	}
1619 	spin_unlock_bh(&adap->tid_release_lock);
1620 }
1621 
1622 /*
1623  * Process the list of pending TID release requests.
1624  */
1625 static void process_tid_release_list(struct work_struct *work)
1626 {
1627 	struct sk_buff *skb;
1628 	struct adapter *adap;
1629 
1630 	adap = container_of(work, struct adapter, tid_release_task);
1631 
1632 	spin_lock_bh(&adap->tid_release_lock);
1633 	while (adap->tid_release_head) {
1634 		void **p = adap->tid_release_head;
1635 		unsigned int chan = (uintptr_t)p & 3;
1636 		p = (void *)p - chan;
1637 
1638 		adap->tid_release_head = *p;
1639 		*p = NULL;
1640 		spin_unlock_bh(&adap->tid_release_lock);
1641 
1642 		while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1643 					 GFP_KERNEL)))
1644 			schedule_timeout_uninterruptible(1);
1645 
1646 		mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1647 		t4_ofld_send(adap, skb);
1648 		spin_lock_bh(&adap->tid_release_lock);
1649 	}
1650 	adap->tid_release_task_busy = false;
1651 	spin_unlock_bh(&adap->tid_release_lock);
1652 }
1653 
1654 /*
1655  * Release a TID and inform HW.  If we are unable to allocate the release
1656  * message we defer to a work queue.
1657  */
1658 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
1659 {
1660 	void *old;
1661 	struct sk_buff *skb;
1662 	struct adapter *adap = container_of(t, struct adapter, tids);
1663 
1664 	old = t->tid_tab[tid];
1665 	skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1666 	if (likely(skb)) {
1667 		t->tid_tab[tid] = NULL;
1668 		mk_tid_release(skb, chan, tid);
1669 		t4_ofld_send(adap, skb);
1670 	} else
1671 		cxgb4_queue_tid_release(t, chan, tid);
1672 	if (old)
1673 		atomic_dec(&t->tids_in_use);
1674 }
1675 EXPORT_SYMBOL(cxgb4_remove_tid);
1676 
1677 /*
1678  * Allocate and initialize the TID tables.  Returns 0 on success.
1679  */
1680 static int tid_init(struct tid_info *t)
1681 {
1682 	size_t size;
1683 	unsigned int stid_bmap_size;
1684 	unsigned int natids = t->natids;
1685 	struct adapter *adap = container_of(t, struct adapter, tids);
1686 
1687 	stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1688 	size = t->ntids * sizeof(*t->tid_tab) +
1689 	       natids * sizeof(*t->atid_tab) +
1690 	       t->nstids * sizeof(*t->stid_tab) +
1691 	       t->nsftids * sizeof(*t->stid_tab) +
1692 	       stid_bmap_size * sizeof(long) +
1693 	       t->nftids * sizeof(*t->ftid_tab) +
1694 	       t->nsftids * sizeof(*t->ftid_tab);
1695 
1696 	t->tid_tab = t4_alloc_mem(size);
1697 	if (!t->tid_tab)
1698 		return -ENOMEM;
1699 
1700 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1701 	t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1702 	t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1703 	t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1704 	spin_lock_init(&t->stid_lock);
1705 	spin_lock_init(&t->atid_lock);
1706 
1707 	t->stids_in_use = 0;
1708 	t->afree = NULL;
1709 	t->atids_in_use = 0;
1710 	atomic_set(&t->tids_in_use, 0);
1711 
1712 	/* Setup the free list for atid_tab and clear the stid bitmap. */
1713 	if (natids) {
1714 		while (--natids)
1715 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1716 		t->afree = t->atid_tab;
1717 	}
1718 	bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1719 	/* Reserve stid 0 for T4/T5 adapters */
1720 	if (!t->stid_base &&
1721 	    (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5))
1722 		__set_bit(0, t->stid_bmap);
1723 
1724 	return 0;
1725 }
1726 
1727 /**
1728  *	cxgb4_create_server - create an IP server
1729  *	@dev: the device
1730  *	@stid: the server TID
1731  *	@sip: local IP address to bind server to
1732  *	@sport: the server's TCP port
1733  *	@queue: queue to direct messages from this server to
1734  *
1735  *	Create an IP server for the given port and address.
1736  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
1737  */
1738 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1739 			__be32 sip, __be16 sport, __be16 vlan,
1740 			unsigned int queue)
1741 {
1742 	unsigned int chan;
1743 	struct sk_buff *skb;
1744 	struct adapter *adap;
1745 	struct cpl_pass_open_req *req;
1746 	int ret;
1747 
1748 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1749 	if (!skb)
1750 		return -ENOMEM;
1751 
1752 	adap = netdev2adap(dev);
1753 	req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
1754 	INIT_TP_WR(req, 0);
1755 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1756 	req->local_port = sport;
1757 	req->peer_port = htons(0);
1758 	req->local_ip = sip;
1759 	req->peer_ip = htonl(0);
1760 	chan = rxq_to_chan(&adap->sge, queue);
1761 	req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1762 	req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1763 				SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1764 	ret = t4_mgmt_tx(adap, skb);
1765 	return net_xmit_eval(ret);
1766 }
1767 EXPORT_SYMBOL(cxgb4_create_server);
1768 
1769 /*	cxgb4_create_server6 - create an IPv6 server
1770  *	@dev: the device
1771  *	@stid: the server TID
1772  *	@sip: local IPv6 address to bind server to
1773  *	@sport: the server's TCP port
1774  *	@queue: queue to direct messages from this server to
1775  *
1776  *	Create an IPv6 server for the given port and address.
1777  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
1778  */
1779 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1780 			 const struct in6_addr *sip, __be16 sport,
1781 			 unsigned int queue)
1782 {
1783 	unsigned int chan;
1784 	struct sk_buff *skb;
1785 	struct adapter *adap;
1786 	struct cpl_pass_open_req6 *req;
1787 	int ret;
1788 
1789 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1790 	if (!skb)
1791 		return -ENOMEM;
1792 
1793 	adap = netdev2adap(dev);
1794 	req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
1795 	INIT_TP_WR(req, 0);
1796 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1797 	req->local_port = sport;
1798 	req->peer_port = htons(0);
1799 	req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1800 	req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1801 	req->peer_ip_hi = cpu_to_be64(0);
1802 	req->peer_ip_lo = cpu_to_be64(0);
1803 	chan = rxq_to_chan(&adap->sge, queue);
1804 	req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1805 	req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1806 				SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1807 	ret = t4_mgmt_tx(adap, skb);
1808 	return net_xmit_eval(ret);
1809 }
1810 EXPORT_SYMBOL(cxgb4_create_server6);
1811 
1812 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1813 			unsigned int queue, bool ipv6)
1814 {
1815 	struct sk_buff *skb;
1816 	struct adapter *adap;
1817 	struct cpl_close_listsvr_req *req;
1818 	int ret;
1819 
1820 	adap = netdev2adap(dev);
1821 
1822 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1823 	if (!skb)
1824 		return -ENOMEM;
1825 
1826 	req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
1827 	INIT_TP_WR(req, 0);
1828 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1829 	req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1830 				LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1831 	ret = t4_mgmt_tx(adap, skb);
1832 	return net_xmit_eval(ret);
1833 }
1834 EXPORT_SYMBOL(cxgb4_remove_server);
1835 
1836 /**
1837  *	cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1838  *	@mtus: the HW MTU table
1839  *	@mtu: the target MTU
1840  *	@idx: index of selected entry in the MTU table
1841  *
1842  *	Returns the index and the value in the HW MTU table that is closest to
1843  *	but does not exceed @mtu, unless @mtu is smaller than any value in the
1844  *	table, in which case that smallest available value is selected.
1845  */
1846 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1847 			    unsigned int *idx)
1848 {
1849 	unsigned int i = 0;
1850 
1851 	while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1852 		++i;
1853 	if (idx)
1854 		*idx = i;
1855 	return mtus[i];
1856 }
1857 EXPORT_SYMBOL(cxgb4_best_mtu);
1858 
1859 /**
1860  *     cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1861  *     @mtus: the HW MTU table
1862  *     @header_size: Header Size
1863  *     @data_size_max: maximum Data Segment Size
1864  *     @data_size_align: desired Data Segment Size Alignment (2^N)
1865  *     @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1866  *
1867  *     Similar to cxgb4_best_mtu() but instead of searching the Hardware
1868  *     MTU Table based solely on a Maximum MTU parameter, we break that
1869  *     parameter up into a Header Size and Maximum Data Segment Size, and
1870  *     provide a desired Data Segment Size Alignment.  If we find an MTU in
1871  *     the Hardware MTU Table which will result in a Data Segment Size with
1872  *     the requested alignment _and_ that MTU isn't "too far" from the
1873  *     closest MTU, then we'll return that rather than the closest MTU.
1874  */
1875 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1876 				    unsigned short header_size,
1877 				    unsigned short data_size_max,
1878 				    unsigned short data_size_align,
1879 				    unsigned int *mtu_idxp)
1880 {
1881 	unsigned short max_mtu = header_size + data_size_max;
1882 	unsigned short data_size_align_mask = data_size_align - 1;
1883 	int mtu_idx, aligned_mtu_idx;
1884 
1885 	/* Scan the MTU Table till we find an MTU which is larger than our
1886 	 * Maximum MTU or we reach the end of the table.  Along the way,
1887 	 * record the last MTU found, if any, which will result in a Data
1888 	 * Segment Length matching the requested alignment.
1889 	 */
1890 	for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1891 		unsigned short data_size = mtus[mtu_idx] - header_size;
1892 
1893 		/* If this MTU minus the Header Size would result in a
1894 		 * Data Segment Size of the desired alignment, remember it.
1895 		 */
1896 		if ((data_size & data_size_align_mask) == 0)
1897 			aligned_mtu_idx = mtu_idx;
1898 
1899 		/* If we're not at the end of the Hardware MTU Table and the
1900 		 * next element is larger than our Maximum MTU, drop out of
1901 		 * the loop.
1902 		 */
1903 		if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1904 			break;
1905 	}
1906 
1907 	/* If we fell out of the loop because we ran to the end of the table,
1908 	 * then we just have to use the last [largest] entry.
1909 	 */
1910 	if (mtu_idx == NMTUS)
1911 		mtu_idx--;
1912 
1913 	/* If we found an MTU which resulted in the requested Data Segment
1914 	 * Length alignment and that's "not far" from the largest MTU which is
1915 	 * less than or equal to the maximum MTU, then use that.
1916 	 */
1917 	if (aligned_mtu_idx >= 0 &&
1918 	    mtu_idx - aligned_mtu_idx <= 1)
1919 		mtu_idx = aligned_mtu_idx;
1920 
1921 	/* If the caller has passed in an MTU Index pointer, pass the
1922 	 * MTU Index back.  Return the MTU value.
1923 	 */
1924 	if (mtu_idxp)
1925 		*mtu_idxp = mtu_idx;
1926 	return mtus[mtu_idx];
1927 }
1928 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1929 
1930 /**
1931  *	cxgb4_port_chan - get the HW channel of a port
1932  *	@dev: the net device for the port
1933  *
1934  *	Return the HW Tx channel of the given port.
1935  */
1936 unsigned int cxgb4_port_chan(const struct net_device *dev)
1937 {
1938 	return netdev2pinfo(dev)->tx_chan;
1939 }
1940 EXPORT_SYMBOL(cxgb4_port_chan);
1941 
1942 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1943 {
1944 	struct adapter *adap = netdev2adap(dev);
1945 	u32 v1, v2, lp_count, hp_count;
1946 
1947 	v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1948 	v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1949 	if (is_t4(adap->params.chip)) {
1950 		lp_count = LP_COUNT_G(v1);
1951 		hp_count = HP_COUNT_G(v1);
1952 	} else {
1953 		lp_count = LP_COUNT_T5_G(v1);
1954 		hp_count = HP_COUNT_T5_G(v2);
1955 	}
1956 	return lpfifo ? lp_count : hp_count;
1957 }
1958 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1959 
1960 /**
1961  *	cxgb4_port_viid - get the VI id of a port
1962  *	@dev: the net device for the port
1963  *
1964  *	Return the VI id of the given port.
1965  */
1966 unsigned int cxgb4_port_viid(const struct net_device *dev)
1967 {
1968 	return netdev2pinfo(dev)->viid;
1969 }
1970 EXPORT_SYMBOL(cxgb4_port_viid);
1971 
1972 /**
1973  *	cxgb4_port_idx - get the index of a port
1974  *	@dev: the net device for the port
1975  *
1976  *	Return the index of the given port.
1977  */
1978 unsigned int cxgb4_port_idx(const struct net_device *dev)
1979 {
1980 	return netdev2pinfo(dev)->port_id;
1981 }
1982 EXPORT_SYMBOL(cxgb4_port_idx);
1983 
1984 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1985 			 struct tp_tcp_stats *v6)
1986 {
1987 	struct adapter *adap = pci_get_drvdata(pdev);
1988 
1989 	spin_lock(&adap->stats_lock);
1990 	t4_tp_get_tcp_stats(adap, v4, v6);
1991 	spin_unlock(&adap->stats_lock);
1992 }
1993 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1994 
1995 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1996 		      const unsigned int *pgsz_order)
1997 {
1998 	struct adapter *adap = netdev2adap(dev);
1999 
2000 	t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
2001 	t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
2002 		     HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
2003 		     HPZ3_V(pgsz_order[3]));
2004 }
2005 EXPORT_SYMBOL(cxgb4_iscsi_init);
2006 
2007 int cxgb4_flush_eq_cache(struct net_device *dev)
2008 {
2009 	struct adapter *adap = netdev2adap(dev);
2010 
2011 	return t4_sge_ctxt_flush(adap, adap->mbox);
2012 }
2013 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
2014 
2015 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
2016 {
2017 	u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
2018 	__be64 indices;
2019 	int ret;
2020 
2021 	spin_lock(&adap->win0_lock);
2022 	ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
2023 			   sizeof(indices), (__be32 *)&indices,
2024 			   T4_MEMORY_READ);
2025 	spin_unlock(&adap->win0_lock);
2026 	if (!ret) {
2027 		*cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
2028 		*pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
2029 	}
2030 	return ret;
2031 }
2032 
2033 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
2034 			u16 size)
2035 {
2036 	struct adapter *adap = netdev2adap(dev);
2037 	u16 hw_pidx, hw_cidx;
2038 	int ret;
2039 
2040 	ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
2041 	if (ret)
2042 		goto out;
2043 
2044 	if (pidx != hw_pidx) {
2045 		u16 delta;
2046 		u32 val;
2047 
2048 		if (pidx >= hw_pidx)
2049 			delta = pidx - hw_pidx;
2050 		else
2051 			delta = size - hw_pidx + pidx;
2052 
2053 		if (is_t4(adap->params.chip))
2054 			val = PIDX_V(delta);
2055 		else
2056 			val = PIDX_T5_V(delta);
2057 		wmb();
2058 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2059 			     QID_V(qid) | val);
2060 	}
2061 out:
2062 	return ret;
2063 }
2064 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
2065 
2066 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
2067 {
2068 	struct adapter *adap;
2069 	u32 offset, memtype, memaddr;
2070 	u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
2071 	u32 edc0_end, edc1_end, mc0_end, mc1_end;
2072 	int ret;
2073 
2074 	adap = netdev2adap(dev);
2075 
2076 	offset = ((stag >> 8) * 32) + adap->vres.stag.start;
2077 
2078 	/* Figure out where the offset lands in the Memory Type/Address scheme.
2079 	 * This code assumes that the memory is laid out starting at offset 0
2080 	 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
2081 	 * and EDC1.  Some cards will have neither MC0 nor MC1, most cards have
2082 	 * MC0, and some have both MC0 and MC1.
2083 	 */
2084 	size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
2085 	edc0_size = EDRAM0_SIZE_G(size) << 20;
2086 	size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
2087 	edc1_size = EDRAM1_SIZE_G(size) << 20;
2088 	size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
2089 	mc0_size = EXT_MEM0_SIZE_G(size) << 20;
2090 
2091 	edc0_end = edc0_size;
2092 	edc1_end = edc0_end + edc1_size;
2093 	mc0_end = edc1_end + mc0_size;
2094 
2095 	if (offset < edc0_end) {
2096 		memtype = MEM_EDC0;
2097 		memaddr = offset;
2098 	} else if (offset < edc1_end) {
2099 		memtype = MEM_EDC1;
2100 		memaddr = offset - edc0_end;
2101 	} else {
2102 		if (offset < mc0_end) {
2103 			memtype = MEM_MC0;
2104 			memaddr = offset - edc1_end;
2105 		} else if (is_t5(adap->params.chip)) {
2106 			size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
2107 			mc1_size = EXT_MEM1_SIZE_G(size) << 20;
2108 			mc1_end = mc0_end + mc1_size;
2109 			if (offset < mc1_end) {
2110 				memtype = MEM_MC1;
2111 				memaddr = offset - mc0_end;
2112 			} else {
2113 				/* offset beyond the end of any memory */
2114 				goto err;
2115 			}
2116 		} else {
2117 			/* T4/T6 only has a single memory channel */
2118 			goto err;
2119 		}
2120 	}
2121 
2122 	spin_lock(&adap->win0_lock);
2123 	ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
2124 	spin_unlock(&adap->win0_lock);
2125 	return ret;
2126 
2127 err:
2128 	dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
2129 		stag, offset);
2130 	return -EINVAL;
2131 }
2132 EXPORT_SYMBOL(cxgb4_read_tpte);
2133 
2134 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
2135 {
2136 	u32 hi, lo;
2137 	struct adapter *adap;
2138 
2139 	adap = netdev2adap(dev);
2140 	lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
2141 	hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
2142 
2143 	return ((u64)hi << 32) | (u64)lo;
2144 }
2145 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
2146 
2147 int cxgb4_bar2_sge_qregs(struct net_device *dev,
2148 			 unsigned int qid,
2149 			 enum cxgb4_bar2_qtype qtype,
2150 			 int user,
2151 			 u64 *pbar2_qoffset,
2152 			 unsigned int *pbar2_qid)
2153 {
2154 	return t4_bar2_sge_qregs(netdev2adap(dev),
2155 				 qid,
2156 				 (qtype == CXGB4_BAR2_QTYPE_EGRESS
2157 				  ? T4_BAR2_QTYPE_EGRESS
2158 				  : T4_BAR2_QTYPE_INGRESS),
2159 				 user,
2160 				 pbar2_qoffset,
2161 				 pbar2_qid);
2162 }
2163 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
2164 
2165 static struct pci_driver cxgb4_driver;
2166 
2167 static void check_neigh_update(struct neighbour *neigh)
2168 {
2169 	const struct device *parent;
2170 	const struct net_device *netdev = neigh->dev;
2171 
2172 	if (netdev->priv_flags & IFF_802_1Q_VLAN)
2173 		netdev = vlan_dev_real_dev(netdev);
2174 	parent = netdev->dev.parent;
2175 	if (parent && parent->driver == &cxgb4_driver.driver)
2176 		t4_l2t_update(dev_get_drvdata(parent), neigh);
2177 }
2178 
2179 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2180 		       void *data)
2181 {
2182 	switch (event) {
2183 	case NETEVENT_NEIGH_UPDATE:
2184 		check_neigh_update(data);
2185 		break;
2186 	case NETEVENT_REDIRECT:
2187 	default:
2188 		break;
2189 	}
2190 	return 0;
2191 }
2192 
2193 static bool netevent_registered;
2194 static struct notifier_block cxgb4_netevent_nb = {
2195 	.notifier_call = netevent_cb
2196 };
2197 
2198 static void drain_db_fifo(struct adapter *adap, int usecs)
2199 {
2200 	u32 v1, v2, lp_count, hp_count;
2201 
2202 	do {
2203 		v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
2204 		v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
2205 		if (is_t4(adap->params.chip)) {
2206 			lp_count = LP_COUNT_G(v1);
2207 			hp_count = HP_COUNT_G(v1);
2208 		} else {
2209 			lp_count = LP_COUNT_T5_G(v1);
2210 			hp_count = HP_COUNT_T5_G(v2);
2211 		}
2212 
2213 		if (lp_count == 0 && hp_count == 0)
2214 			break;
2215 		set_current_state(TASK_UNINTERRUPTIBLE);
2216 		schedule_timeout(usecs_to_jiffies(usecs));
2217 	} while (1);
2218 }
2219 
2220 static void disable_txq_db(struct sge_txq *q)
2221 {
2222 	unsigned long flags;
2223 
2224 	spin_lock_irqsave(&q->db_lock, flags);
2225 	q->db_disabled = 1;
2226 	spin_unlock_irqrestore(&q->db_lock, flags);
2227 }
2228 
2229 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
2230 {
2231 	spin_lock_irq(&q->db_lock);
2232 	if (q->db_pidx_inc) {
2233 		/* Make sure that all writes to the TX descriptors
2234 		 * are committed before we tell HW about them.
2235 		 */
2236 		wmb();
2237 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2238 			     QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
2239 		q->db_pidx_inc = 0;
2240 	}
2241 	q->db_disabled = 0;
2242 	spin_unlock_irq(&q->db_lock);
2243 }
2244 
2245 static void disable_dbs(struct adapter *adap)
2246 {
2247 	int i;
2248 
2249 	for_each_ethrxq(&adap->sge, i)
2250 		disable_txq_db(&adap->sge.ethtxq[i].q);
2251 	for_each_ofldrxq(&adap->sge, i)
2252 		disable_txq_db(&adap->sge.ofldtxq[i].q);
2253 	for_each_port(adap, i)
2254 		disable_txq_db(&adap->sge.ctrlq[i].q);
2255 }
2256 
2257 static void enable_dbs(struct adapter *adap)
2258 {
2259 	int i;
2260 
2261 	for_each_ethrxq(&adap->sge, i)
2262 		enable_txq_db(adap, &adap->sge.ethtxq[i].q);
2263 	for_each_ofldrxq(&adap->sge, i)
2264 		enable_txq_db(adap, &adap->sge.ofldtxq[i].q);
2265 	for_each_port(adap, i)
2266 		enable_txq_db(adap, &adap->sge.ctrlq[i].q);
2267 }
2268 
2269 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
2270 {
2271 	if (adap->uld_handle[CXGB4_ULD_RDMA])
2272 		ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
2273 				cmd);
2274 }
2275 
2276 static void process_db_full(struct work_struct *work)
2277 {
2278 	struct adapter *adap;
2279 
2280 	adap = container_of(work, struct adapter, db_full_task);
2281 
2282 	drain_db_fifo(adap, dbfifo_drain_delay);
2283 	enable_dbs(adap);
2284 	notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2285 	if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2286 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2287 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
2288 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
2289 	else
2290 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2291 				 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
2292 }
2293 
2294 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2295 {
2296 	u16 hw_pidx, hw_cidx;
2297 	int ret;
2298 
2299 	spin_lock_irq(&q->db_lock);
2300 	ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2301 	if (ret)
2302 		goto out;
2303 	if (q->db_pidx != hw_pidx) {
2304 		u16 delta;
2305 		u32 val;
2306 
2307 		if (q->db_pidx >= hw_pidx)
2308 			delta = q->db_pidx - hw_pidx;
2309 		else
2310 			delta = q->size - hw_pidx + q->db_pidx;
2311 
2312 		if (is_t4(adap->params.chip))
2313 			val = PIDX_V(delta);
2314 		else
2315 			val = PIDX_T5_V(delta);
2316 		wmb();
2317 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2318 			     QID_V(q->cntxt_id) | val);
2319 	}
2320 out:
2321 	q->db_disabled = 0;
2322 	q->db_pidx_inc = 0;
2323 	spin_unlock_irq(&q->db_lock);
2324 	if (ret)
2325 		CH_WARN(adap, "DB drop recovery failed.\n");
2326 }
2327 static void recover_all_queues(struct adapter *adap)
2328 {
2329 	int i;
2330 
2331 	for_each_ethrxq(&adap->sge, i)
2332 		sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2333 	for_each_ofldrxq(&adap->sge, i)
2334 		sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
2335 	for_each_port(adap, i)
2336 		sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2337 }
2338 
2339 static void process_db_drop(struct work_struct *work)
2340 {
2341 	struct adapter *adap;
2342 
2343 	adap = container_of(work, struct adapter, db_drop_task);
2344 
2345 	if (is_t4(adap->params.chip)) {
2346 		drain_db_fifo(adap, dbfifo_drain_delay);
2347 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2348 		drain_db_fifo(adap, dbfifo_drain_delay);
2349 		recover_all_queues(adap);
2350 		drain_db_fifo(adap, dbfifo_drain_delay);
2351 		enable_dbs(adap);
2352 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2353 	} else if (is_t5(adap->params.chip)) {
2354 		u32 dropped_db = t4_read_reg(adap, 0x010ac);
2355 		u16 qid = (dropped_db >> 15) & 0x1ffff;
2356 		u16 pidx_inc = dropped_db & 0x1fff;
2357 		u64 bar2_qoffset;
2358 		unsigned int bar2_qid;
2359 		int ret;
2360 
2361 		ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2362 					0, &bar2_qoffset, &bar2_qid);
2363 		if (ret)
2364 			dev_err(adap->pdev_dev, "doorbell drop recovery: "
2365 				"qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2366 		else
2367 			writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2368 			       adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2369 
2370 		/* Re-enable BAR2 WC */
2371 		t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2372 	}
2373 
2374 	if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2375 		t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2376 }
2377 
2378 void t4_db_full(struct adapter *adap)
2379 {
2380 	if (is_t4(adap->params.chip)) {
2381 		disable_dbs(adap);
2382 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2383 		t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2384 				 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2385 		queue_work(adap->workq, &adap->db_full_task);
2386 	}
2387 }
2388 
2389 void t4_db_dropped(struct adapter *adap)
2390 {
2391 	if (is_t4(adap->params.chip)) {
2392 		disable_dbs(adap);
2393 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2394 	}
2395 	queue_work(adap->workq, &adap->db_drop_task);
2396 }
2397 
2398 static void uld_attach(struct adapter *adap, unsigned int uld)
2399 {
2400 	void *handle;
2401 	struct cxgb4_lld_info lli;
2402 	unsigned short i;
2403 
2404 	lli.pdev = adap->pdev;
2405 	lli.pf = adap->pf;
2406 	lli.l2t = adap->l2t;
2407 	lli.tids = &adap->tids;
2408 	lli.ports = adap->port;
2409 	lli.vr = &adap->vres;
2410 	lli.mtus = adap->params.mtus;
2411 	if (uld == CXGB4_ULD_RDMA) {
2412 		lli.rxq_ids = adap->sge.rdma_rxq;
2413 		lli.ciq_ids = adap->sge.rdma_ciq;
2414 		lli.nrxq = adap->sge.rdmaqs;
2415 		lli.nciq = adap->sge.rdmaciqs;
2416 	} else if (uld == CXGB4_ULD_ISCSI) {
2417 		lli.rxq_ids = adap->sge.ofld_rxq;
2418 		lli.nrxq = adap->sge.ofldqsets;
2419 	}
2420 	lli.ntxq = adap->sge.ofldqsets;
2421 	lli.nchan = adap->params.nports;
2422 	lli.nports = adap->params.nports;
2423 	lli.wr_cred = adap->params.ofldq_wr_cred;
2424 	lli.adapter_type = adap->params.chip;
2425 	lli.iscsi_iolen = MAXRXDATA_G(t4_read_reg(adap, TP_PARA_REG2_A));
2426 	lli.cclk_ps = 1000000000 / adap->params.vpd.cclk;
2427 	lli.udb_density = 1 << adap->params.sge.eq_qpp;
2428 	lli.ucq_density = 1 << adap->params.sge.iq_qpp;
2429 	lli.filt_mode = adap->params.tp.vlan_pri_map;
2430 	/* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
2431 	for (i = 0; i < NCHAN; i++)
2432 		lli.tx_modq[i] = i;
2433 	lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS_A);
2434 	lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL_A);
2435 	lli.fw_vers = adap->params.fw_vers;
2436 	lli.dbfifo_int_thresh = dbfifo_int_thresh;
2437 	lli.sge_ingpadboundary = adap->sge.fl_align;
2438 	lli.sge_egrstatuspagesize = adap->sge.stat_len;
2439 	lli.sge_pktshift = adap->sge.pktshift;
2440 	lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
2441 	lli.max_ordird_qp = adap->params.max_ordird_qp;
2442 	lli.max_ird_adapter = adap->params.max_ird_adapter;
2443 	lli.ulptx_memwrite_dsgl = adap->params.ulptx_memwrite_dsgl;
2444 	lli.nodeid = dev_to_node(adap->pdev_dev);
2445 
2446 	handle = ulds[uld].add(&lli);
2447 	if (IS_ERR(handle)) {
2448 		dev_warn(adap->pdev_dev,
2449 			 "could not attach to the %s driver, error %ld\n",
2450 			 uld_str[uld], PTR_ERR(handle));
2451 		return;
2452 	}
2453 
2454 	adap->uld_handle[uld] = handle;
2455 
2456 	if (!netevent_registered) {
2457 		register_netevent_notifier(&cxgb4_netevent_nb);
2458 		netevent_registered = true;
2459 	}
2460 
2461 	if (adap->flags & FULL_INIT_DONE)
2462 		ulds[uld].state_change(handle, CXGB4_STATE_UP);
2463 }
2464 
2465 static void attach_ulds(struct adapter *adap)
2466 {
2467 	unsigned int i;
2468 
2469 	spin_lock(&adap_rcu_lock);
2470 	list_add_tail_rcu(&adap->rcu_node, &adap_rcu_list);
2471 	spin_unlock(&adap_rcu_lock);
2472 
2473 	mutex_lock(&uld_mutex);
2474 	list_add_tail(&adap->list_node, &adapter_list);
2475 	for (i = 0; i < CXGB4_ULD_MAX; i++)
2476 		if (ulds[i].add)
2477 			uld_attach(adap, i);
2478 	mutex_unlock(&uld_mutex);
2479 }
2480 
2481 static void detach_ulds(struct adapter *adap)
2482 {
2483 	unsigned int i;
2484 
2485 	mutex_lock(&uld_mutex);
2486 	list_del(&adap->list_node);
2487 	for (i = 0; i < CXGB4_ULD_MAX; i++)
2488 		if (adap->uld_handle[i]) {
2489 			ulds[i].state_change(adap->uld_handle[i],
2490 					     CXGB4_STATE_DETACH);
2491 			adap->uld_handle[i] = NULL;
2492 		}
2493 	if (netevent_registered && list_empty(&adapter_list)) {
2494 		unregister_netevent_notifier(&cxgb4_netevent_nb);
2495 		netevent_registered = false;
2496 	}
2497 	mutex_unlock(&uld_mutex);
2498 
2499 	spin_lock(&adap_rcu_lock);
2500 	list_del_rcu(&adap->rcu_node);
2501 	spin_unlock(&adap_rcu_lock);
2502 }
2503 
2504 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2505 {
2506 	unsigned int i;
2507 
2508 	mutex_lock(&uld_mutex);
2509 	for (i = 0; i < CXGB4_ULD_MAX; i++)
2510 		if (adap->uld_handle[i])
2511 			ulds[i].state_change(adap->uld_handle[i], new_state);
2512 	mutex_unlock(&uld_mutex);
2513 }
2514 
2515 /**
2516  *	cxgb4_register_uld - register an upper-layer driver
2517  *	@type: the ULD type
2518  *	@p: the ULD methods
2519  *
2520  *	Registers an upper-layer driver with this driver and notifies the ULD
2521  *	about any presently available devices that support its type.  Returns
2522  *	%-EBUSY if a ULD of the same type is already registered.
2523  */
2524 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
2525 {
2526 	int ret = 0;
2527 	struct adapter *adap;
2528 
2529 	if (type >= CXGB4_ULD_MAX)
2530 		return -EINVAL;
2531 	mutex_lock(&uld_mutex);
2532 	if (ulds[type].add) {
2533 		ret = -EBUSY;
2534 		goto out;
2535 	}
2536 	ulds[type] = *p;
2537 	list_for_each_entry(adap, &adapter_list, list_node)
2538 		uld_attach(adap, type);
2539 out:	mutex_unlock(&uld_mutex);
2540 	return ret;
2541 }
2542 EXPORT_SYMBOL(cxgb4_register_uld);
2543 
2544 /**
2545  *	cxgb4_unregister_uld - unregister an upper-layer driver
2546  *	@type: the ULD type
2547  *
2548  *	Unregisters an existing upper-layer driver.
2549  */
2550 int cxgb4_unregister_uld(enum cxgb4_uld type)
2551 {
2552 	struct adapter *adap;
2553 
2554 	if (type >= CXGB4_ULD_MAX)
2555 		return -EINVAL;
2556 	mutex_lock(&uld_mutex);
2557 	list_for_each_entry(adap, &adapter_list, list_node)
2558 		adap->uld_handle[type] = NULL;
2559 	ulds[type].add = NULL;
2560 	mutex_unlock(&uld_mutex);
2561 	return 0;
2562 }
2563 EXPORT_SYMBOL(cxgb4_unregister_uld);
2564 
2565 #if IS_ENABLED(CONFIG_IPV6)
2566 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2567 				   unsigned long event, void *data)
2568 {
2569 	struct inet6_ifaddr *ifa = data;
2570 	struct net_device *event_dev = ifa->idev->dev;
2571 	const struct device *parent = NULL;
2572 #if IS_ENABLED(CONFIG_BONDING)
2573 	struct adapter *adap;
2574 #endif
2575 	if (event_dev->priv_flags & IFF_802_1Q_VLAN)
2576 		event_dev = vlan_dev_real_dev(event_dev);
2577 #if IS_ENABLED(CONFIG_BONDING)
2578 	if (event_dev->flags & IFF_MASTER) {
2579 		list_for_each_entry(adap, &adapter_list, list_node) {
2580 			switch (event) {
2581 			case NETDEV_UP:
2582 				cxgb4_clip_get(adap->port[0],
2583 					       (const u32 *)ifa, 1);
2584 				break;
2585 			case NETDEV_DOWN:
2586 				cxgb4_clip_release(adap->port[0],
2587 						   (const u32 *)ifa, 1);
2588 				break;
2589 			default:
2590 				break;
2591 			}
2592 		}
2593 		return NOTIFY_OK;
2594 	}
2595 #endif
2596 
2597 	if (event_dev)
2598 		parent = event_dev->dev.parent;
2599 
2600 	if (parent && parent->driver == &cxgb4_driver.driver) {
2601 		switch (event) {
2602 		case NETDEV_UP:
2603 			cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2604 			break;
2605 		case NETDEV_DOWN:
2606 			cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2607 			break;
2608 		default:
2609 			break;
2610 		}
2611 	}
2612 	return NOTIFY_OK;
2613 }
2614 
2615 static bool inet6addr_registered;
2616 static struct notifier_block cxgb4_inet6addr_notifier = {
2617 	.notifier_call = cxgb4_inet6addr_handler
2618 };
2619 
2620 static void update_clip(const struct adapter *adap)
2621 {
2622 	int i;
2623 	struct net_device *dev;
2624 	int ret;
2625 
2626 	rcu_read_lock();
2627 
2628 	for (i = 0; i < MAX_NPORTS; i++) {
2629 		dev = adap->port[i];
2630 		ret = 0;
2631 
2632 		if (dev)
2633 			ret = cxgb4_update_root_dev_clip(dev);
2634 
2635 		if (ret < 0)
2636 			break;
2637 	}
2638 	rcu_read_unlock();
2639 }
2640 #endif /* IS_ENABLED(CONFIG_IPV6) */
2641 
2642 /**
2643  *	cxgb_up - enable the adapter
2644  *	@adap: adapter being enabled
2645  *
2646  *	Called when the first port is enabled, this function performs the
2647  *	actions necessary to make an adapter operational, such as completing
2648  *	the initialization of HW modules, and enabling interrupts.
2649  *
2650  *	Must be called with the rtnl lock held.
2651  */
2652 static int cxgb_up(struct adapter *adap)
2653 {
2654 	int err;
2655 
2656 	err = setup_sge_queues(adap);
2657 	if (err)
2658 		goto out;
2659 	err = setup_rss(adap);
2660 	if (err)
2661 		goto freeq;
2662 
2663 	if (adap->flags & USING_MSIX) {
2664 		name_msix_vecs(adap);
2665 		err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2666 				  adap->msix_info[0].desc, adap);
2667 		if (err)
2668 			goto irq_err;
2669 
2670 		err = request_msix_queue_irqs(adap);
2671 		if (err) {
2672 			free_irq(adap->msix_info[0].vec, adap);
2673 			goto irq_err;
2674 		}
2675 	} else {
2676 		err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2677 				  (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2678 				  adap->port[0]->name, adap);
2679 		if (err)
2680 			goto irq_err;
2681 	}
2682 	enable_rx(adap);
2683 	t4_sge_start(adap);
2684 	t4_intr_enable(adap);
2685 	adap->flags |= FULL_INIT_DONE;
2686 	notify_ulds(adap, CXGB4_STATE_UP);
2687 #if IS_ENABLED(CONFIG_IPV6)
2688 	update_clip(adap);
2689 #endif
2690  out:
2691 	return err;
2692  irq_err:
2693 	dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2694  freeq:
2695 	t4_free_sge_resources(adap);
2696 	goto out;
2697 }
2698 
2699 static void cxgb_down(struct adapter *adapter)
2700 {
2701 	cancel_work_sync(&adapter->tid_release_task);
2702 	cancel_work_sync(&adapter->db_full_task);
2703 	cancel_work_sync(&adapter->db_drop_task);
2704 	adapter->tid_release_task_busy = false;
2705 	adapter->tid_release_head = NULL;
2706 
2707 	t4_sge_stop(adapter);
2708 	t4_free_sge_resources(adapter);
2709 	adapter->flags &= ~FULL_INIT_DONE;
2710 }
2711 
2712 /*
2713  * net_device operations
2714  */
2715 static int cxgb_open(struct net_device *dev)
2716 {
2717 	int err;
2718 	struct port_info *pi = netdev_priv(dev);
2719 	struct adapter *adapter = pi->adapter;
2720 
2721 	netif_carrier_off(dev);
2722 
2723 	if (!(adapter->flags & FULL_INIT_DONE)) {
2724 		err = cxgb_up(adapter);
2725 		if (err < 0)
2726 			return err;
2727 	}
2728 
2729 	err = link_start(dev);
2730 	if (!err)
2731 		netif_tx_start_all_queues(dev);
2732 	return err;
2733 }
2734 
2735 static int cxgb_close(struct net_device *dev)
2736 {
2737 	struct port_info *pi = netdev_priv(dev);
2738 	struct adapter *adapter = pi->adapter;
2739 
2740 	netif_tx_stop_all_queues(dev);
2741 	netif_carrier_off(dev);
2742 	return t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
2743 }
2744 
2745 /* Return an error number if the indicated filter isn't writable ...
2746  */
2747 static int writable_filter(struct filter_entry *f)
2748 {
2749 	if (f->locked)
2750 		return -EPERM;
2751 	if (f->pending)
2752 		return -EBUSY;
2753 
2754 	return 0;
2755 }
2756 
2757 /* Delete the filter at the specified index (if valid).  The checks for all
2758  * the common problems with doing this like the filter being locked, currently
2759  * pending in another operation, etc.
2760  */
2761 static int delete_filter(struct adapter *adapter, unsigned int fidx)
2762 {
2763 	struct filter_entry *f;
2764 	int ret;
2765 
2766 	if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
2767 		return -EINVAL;
2768 
2769 	f = &adapter->tids.ftid_tab[fidx];
2770 	ret = writable_filter(f);
2771 	if (ret)
2772 		return ret;
2773 	if (f->valid)
2774 		return del_filter_wr(adapter, fidx);
2775 
2776 	return 0;
2777 }
2778 
2779 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2780 		__be32 sip, __be16 sport, __be16 vlan,
2781 		unsigned int queue, unsigned char port, unsigned char mask)
2782 {
2783 	int ret;
2784 	struct filter_entry *f;
2785 	struct adapter *adap;
2786 	int i;
2787 	u8 *val;
2788 
2789 	adap = netdev2adap(dev);
2790 
2791 	/* Adjust stid to correct filter index */
2792 	stid -= adap->tids.sftid_base;
2793 	stid += adap->tids.nftids;
2794 
2795 	/* Check to make sure the filter requested is writable ...
2796 	 */
2797 	f = &adap->tids.ftid_tab[stid];
2798 	ret = writable_filter(f);
2799 	if (ret)
2800 		return ret;
2801 
2802 	/* Clear out any old resources being used by the filter before
2803 	 * we start constructing the new filter.
2804 	 */
2805 	if (f->valid)
2806 		clear_filter(adap, f);
2807 
2808 	/* Clear out filter specifications */
2809 	memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2810 	f->fs.val.lport = cpu_to_be16(sport);
2811 	f->fs.mask.lport  = ~0;
2812 	val = (u8 *)&sip;
2813 	if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2814 		for (i = 0; i < 4; i++) {
2815 			f->fs.val.lip[i] = val[i];
2816 			f->fs.mask.lip[i] = ~0;
2817 		}
2818 		if (adap->params.tp.vlan_pri_map & PORT_F) {
2819 			f->fs.val.iport = port;
2820 			f->fs.mask.iport = mask;
2821 		}
2822 	}
2823 
2824 	if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2825 		f->fs.val.proto = IPPROTO_TCP;
2826 		f->fs.mask.proto = ~0;
2827 	}
2828 
2829 	f->fs.dirsteer = 1;
2830 	f->fs.iq = queue;
2831 	/* Mark filter as locked */
2832 	f->locked = 1;
2833 	f->fs.rpttid = 1;
2834 
2835 	ret = set_filter_wr(adap, stid);
2836 	if (ret) {
2837 		clear_filter(adap, f);
2838 		return ret;
2839 	}
2840 
2841 	return 0;
2842 }
2843 EXPORT_SYMBOL(cxgb4_create_server_filter);
2844 
2845 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2846 		unsigned int queue, bool ipv6)
2847 {
2848 	int ret;
2849 	struct filter_entry *f;
2850 	struct adapter *adap;
2851 
2852 	adap = netdev2adap(dev);
2853 
2854 	/* Adjust stid to correct filter index */
2855 	stid -= adap->tids.sftid_base;
2856 	stid += adap->tids.nftids;
2857 
2858 	f = &adap->tids.ftid_tab[stid];
2859 	/* Unlock the filter */
2860 	f->locked = 0;
2861 
2862 	ret = delete_filter(adap, stid);
2863 	if (ret)
2864 		return ret;
2865 
2866 	return 0;
2867 }
2868 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2869 
2870 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
2871 						struct rtnl_link_stats64 *ns)
2872 {
2873 	struct port_stats stats;
2874 	struct port_info *p = netdev_priv(dev);
2875 	struct adapter *adapter = p->adapter;
2876 
2877 	/* Block retrieving statistics during EEH error
2878 	 * recovery. Otherwise, the recovery might fail
2879 	 * and the PCI device will be removed permanently
2880 	 */
2881 	spin_lock(&adapter->stats_lock);
2882 	if (!netif_device_present(dev)) {
2883 		spin_unlock(&adapter->stats_lock);
2884 		return ns;
2885 	}
2886 	t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2887 				 &p->stats_base);
2888 	spin_unlock(&adapter->stats_lock);
2889 
2890 	ns->tx_bytes   = stats.tx_octets;
2891 	ns->tx_packets = stats.tx_frames;
2892 	ns->rx_bytes   = stats.rx_octets;
2893 	ns->rx_packets = stats.rx_frames;
2894 	ns->multicast  = stats.rx_mcast_frames;
2895 
2896 	/* detailed rx_errors */
2897 	ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2898 			       stats.rx_runt;
2899 	ns->rx_over_errors   = 0;
2900 	ns->rx_crc_errors    = stats.rx_fcs_err;
2901 	ns->rx_frame_errors  = stats.rx_symbol_err;
2902 	ns->rx_fifo_errors   = stats.rx_ovflow0 + stats.rx_ovflow1 +
2903 			       stats.rx_ovflow2 + stats.rx_ovflow3 +
2904 			       stats.rx_trunc0 + stats.rx_trunc1 +
2905 			       stats.rx_trunc2 + stats.rx_trunc3;
2906 	ns->rx_missed_errors = 0;
2907 
2908 	/* detailed tx_errors */
2909 	ns->tx_aborted_errors   = 0;
2910 	ns->tx_carrier_errors   = 0;
2911 	ns->tx_fifo_errors      = 0;
2912 	ns->tx_heartbeat_errors = 0;
2913 	ns->tx_window_errors    = 0;
2914 
2915 	ns->tx_errors = stats.tx_error_frames;
2916 	ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2917 		ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2918 	return ns;
2919 }
2920 
2921 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2922 {
2923 	unsigned int mbox;
2924 	int ret = 0, prtad, devad;
2925 	struct port_info *pi = netdev_priv(dev);
2926 	struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2927 
2928 	switch (cmd) {
2929 	case SIOCGMIIPHY:
2930 		if (pi->mdio_addr < 0)
2931 			return -EOPNOTSUPP;
2932 		data->phy_id = pi->mdio_addr;
2933 		break;
2934 	case SIOCGMIIREG:
2935 	case SIOCSMIIREG:
2936 		if (mdio_phy_id_is_c45(data->phy_id)) {
2937 			prtad = mdio_phy_id_prtad(data->phy_id);
2938 			devad = mdio_phy_id_devad(data->phy_id);
2939 		} else if (data->phy_id < 32) {
2940 			prtad = data->phy_id;
2941 			devad = 0;
2942 			data->reg_num &= 0x1f;
2943 		} else
2944 			return -EINVAL;
2945 
2946 		mbox = pi->adapter->pf;
2947 		if (cmd == SIOCGMIIREG)
2948 			ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2949 					 data->reg_num, &data->val_out);
2950 		else
2951 			ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2952 					 data->reg_num, data->val_in);
2953 		break;
2954 	default:
2955 		return -EOPNOTSUPP;
2956 	}
2957 	return ret;
2958 }
2959 
2960 static void cxgb_set_rxmode(struct net_device *dev)
2961 {
2962 	/* unfortunately we can't return errors to the stack */
2963 	set_rxmode(dev, -1, false);
2964 }
2965 
2966 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2967 {
2968 	int ret;
2969 	struct port_info *pi = netdev_priv(dev);
2970 
2971 	if (new_mtu < 81 || new_mtu > MAX_MTU)         /* accommodate SACK */
2972 		return -EINVAL;
2973 	ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2974 			    -1, -1, -1, true);
2975 	if (!ret)
2976 		dev->mtu = new_mtu;
2977 	return ret;
2978 }
2979 
2980 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2981 {
2982 	int ret;
2983 	struct sockaddr *addr = p;
2984 	struct port_info *pi = netdev_priv(dev);
2985 
2986 	if (!is_valid_ether_addr(addr->sa_data))
2987 		return -EADDRNOTAVAIL;
2988 
2989 	ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2990 			    pi->xact_addr_filt, addr->sa_data, true, true);
2991 	if (ret < 0)
2992 		return ret;
2993 
2994 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2995 	pi->xact_addr_filt = ret;
2996 	return 0;
2997 }
2998 
2999 #ifdef CONFIG_NET_POLL_CONTROLLER
3000 static void cxgb_netpoll(struct net_device *dev)
3001 {
3002 	struct port_info *pi = netdev_priv(dev);
3003 	struct adapter *adap = pi->adapter;
3004 
3005 	if (adap->flags & USING_MSIX) {
3006 		int i;
3007 		struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
3008 
3009 		for (i = pi->nqsets; i; i--, rx++)
3010 			t4_sge_intr_msix(0, &rx->rspq);
3011 	} else
3012 		t4_intr_handler(adap)(0, adap);
3013 }
3014 #endif
3015 
3016 static const struct net_device_ops cxgb4_netdev_ops = {
3017 	.ndo_open             = cxgb_open,
3018 	.ndo_stop             = cxgb_close,
3019 	.ndo_start_xmit       = t4_eth_xmit,
3020 	.ndo_select_queue     =	cxgb_select_queue,
3021 	.ndo_get_stats64      = cxgb_get_stats,
3022 	.ndo_set_rx_mode      = cxgb_set_rxmode,
3023 	.ndo_set_mac_address  = cxgb_set_mac_addr,
3024 	.ndo_set_features     = cxgb_set_features,
3025 	.ndo_validate_addr    = eth_validate_addr,
3026 	.ndo_do_ioctl         = cxgb_ioctl,
3027 	.ndo_change_mtu       = cxgb_change_mtu,
3028 #ifdef CONFIG_NET_POLL_CONTROLLER
3029 	.ndo_poll_controller  = cxgb_netpoll,
3030 #endif
3031 #ifdef CONFIG_CHELSIO_T4_FCOE
3032 	.ndo_fcoe_enable      = cxgb_fcoe_enable,
3033 	.ndo_fcoe_disable     = cxgb_fcoe_disable,
3034 #endif /* CONFIG_CHELSIO_T4_FCOE */
3035 #ifdef CONFIG_NET_RX_BUSY_POLL
3036 	.ndo_busy_poll        = cxgb_busy_poll,
3037 #endif
3038 
3039 };
3040 
3041 void t4_fatal_err(struct adapter *adap)
3042 {
3043 	t4_set_reg_field(adap, SGE_CONTROL_A, GLOBALENABLE_F, 0);
3044 	t4_intr_disable(adap);
3045 	dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3046 }
3047 
3048 static void setup_memwin(struct adapter *adap)
3049 {
3050 	u32 nic_win_base = t4_get_util_window(adap);
3051 
3052 	t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3053 }
3054 
3055 static void setup_memwin_rdma(struct adapter *adap)
3056 {
3057 	if (adap->vres.ocq.size) {
3058 		u32 start;
3059 		unsigned int sz_kb;
3060 
3061 		start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3062 		start &= PCI_BASE_ADDRESS_MEM_MASK;
3063 		start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3064 		sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3065 		t4_write_reg(adap,
3066 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3067 			     start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3068 		t4_write_reg(adap,
3069 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3070 			     adap->vres.ocq.start);
3071 		t4_read_reg(adap,
3072 			    PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3073 	}
3074 }
3075 
3076 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3077 {
3078 	u32 v;
3079 	int ret;
3080 
3081 	/* get device capabilities */
3082 	memset(c, 0, sizeof(*c));
3083 	c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3084 			       FW_CMD_REQUEST_F | FW_CMD_READ_F);
3085 	c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3086 	ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
3087 	if (ret < 0)
3088 		return ret;
3089 
3090 	/* select capabilities we'll be using */
3091 	if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
3092 		if (!vf_acls)
3093 			c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
3094 		else
3095 			c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
3096 	} else if (vf_acls) {
3097 		dev_err(adap->pdev_dev, "virtualization ACLs not supported");
3098 		return ret;
3099 	}
3100 	c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3101 			       FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3102 	ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
3103 	if (ret < 0)
3104 		return ret;
3105 
3106 	ret = t4_config_glbl_rss(adap, adap->pf,
3107 				 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3108 				 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3109 				 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
3110 	if (ret < 0)
3111 		return ret;
3112 
3113 	ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
3114 			  MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3115 			  FW_CMD_CAP_PF);
3116 	if (ret < 0)
3117 		return ret;
3118 
3119 	t4_sge_init(adap);
3120 
3121 	/* tweak some settings */
3122 	t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3123 	t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3124 	t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3125 	v = t4_read_reg(adap, TP_PIO_DATA_A);
3126 	t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3127 
3128 	/* first 4 Tx modulation queues point to consecutive Tx channels */
3129 	adap->params.tp.tx_modq_map = 0xE4;
3130 	t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3131 		     TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3132 
3133 	/* associate each Tx modulation queue with consecutive Tx channels */
3134 	v = 0x84218421;
3135 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3136 			  &v, 1, TP_TX_SCHED_HDR_A);
3137 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3138 			  &v, 1, TP_TX_SCHED_FIFO_A);
3139 	t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3140 			  &v, 1, TP_TX_SCHED_PCMD_A);
3141 
3142 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3143 	if (is_offload(adap)) {
3144 		t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3145 			     TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3146 			     TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3147 			     TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3148 			     TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3149 		t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3150 			     TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3151 			     TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3152 			     TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3153 			     TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3154 	}
3155 
3156 	/* get basic stuff going */
3157 	return t4_early_init(adap, adap->pf);
3158 }
3159 
3160 /*
3161  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
3162  */
3163 #define MAX_ATIDS 8192U
3164 
3165 /*
3166  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3167  *
3168  * If the firmware we're dealing with has Configuration File support, then
3169  * we use that to perform all configuration
3170  */
3171 
3172 /*
3173  * Tweak configuration based on module parameters, etc.  Most of these have
3174  * defaults assigned to them by Firmware Configuration Files (if we're using
3175  * them) but need to be explicitly set if we're using hard-coded
3176  * initialization.  But even in the case of using Firmware Configuration
3177  * Files, we'd like to expose the ability to change these via module
3178  * parameters so these are essentially common tweaks/settings for
3179  * Configuration Files and hard-coded initialization ...
3180  */
3181 static int adap_init0_tweaks(struct adapter *adapter)
3182 {
3183 	/*
3184 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
3185 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
3186 	 * 64B Cache Line Size ...
3187 	 */
3188 	t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3189 
3190 	/*
3191 	 * Process module parameters which affect early initialization.
3192 	 */
3193 	if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3194 		dev_err(&adapter->pdev->dev,
3195 			"Ignoring illegal rx_dma_offset=%d, using 2\n",
3196 			rx_dma_offset);
3197 		rx_dma_offset = 2;
3198 	}
3199 	t4_set_reg_field(adapter, SGE_CONTROL_A,
3200 			 PKTSHIFT_V(PKTSHIFT_M),
3201 			 PKTSHIFT_V(rx_dma_offset));
3202 
3203 	/*
3204 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3205 	 * adds the pseudo header itself.
3206 	 */
3207 	t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3208 			       CSUM_HAS_PSEUDO_HDR_F, 0);
3209 
3210 	return 0;
3211 }
3212 
3213 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3214  * unto themselves and they contain their own firmware to perform their
3215  * tasks ...
3216  */
3217 static int phy_aq1202_version(const u8 *phy_fw_data,
3218 			      size_t phy_fw_size)
3219 {
3220 	int offset;
3221 
3222 	/* At offset 0x8 you're looking for the primary image's
3223 	 * starting offset which is 3 Bytes wide
3224 	 *
3225 	 * At offset 0xa of the primary image, you look for the offset
3226 	 * of the DRAM segment which is 3 Bytes wide.
3227 	 *
3228 	 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3229 	 * wide
3230 	 */
3231 	#define be16(__p) (((__p)[0] << 8) | (__p)[1])
3232 	#define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3233 	#define le24(__p) (le16(__p) | ((__p)[2] << 16))
3234 
3235 	offset = le24(phy_fw_data + 0x8) << 12;
3236 	offset = le24(phy_fw_data + offset + 0xa);
3237 	return be16(phy_fw_data + offset + 0x27e);
3238 
3239 	#undef be16
3240 	#undef le16
3241 	#undef le24
3242 }
3243 
3244 static struct info_10gbt_phy_fw {
3245 	unsigned int phy_fw_id;		/* PCI Device ID */
3246 	char *phy_fw_file;		/* /lib/firmware/ PHY Firmware file */
3247 	int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3248 	int phy_flash;			/* Has FLASH for PHY Firmware */
3249 } phy_info_array[] = {
3250 	{
3251 		PHY_AQ1202_DEVICEID,
3252 		PHY_AQ1202_FIRMWARE,
3253 		phy_aq1202_version,
3254 		1,
3255 	},
3256 	{
3257 		PHY_BCM84834_DEVICEID,
3258 		PHY_BCM84834_FIRMWARE,
3259 		NULL,
3260 		0,
3261 	},
3262 	{ 0, NULL, NULL },
3263 };
3264 
3265 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3266 {
3267 	int i;
3268 
3269 	for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3270 		if (phy_info_array[i].phy_fw_id == devid)
3271 			return &phy_info_array[i];
3272 	}
3273 	return NULL;
3274 }
3275 
3276 /* Handle updating of chip-external 10Gb/s-BT PHY firmware.  This needs to
3277  * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD.  On error
3278  * we return a negative error number.  If we transfer new firmware we return 1
3279  * (from t4_load_phy_fw()).  If we don't do anything we return 0.
3280  */
3281 static int adap_init0_phy(struct adapter *adap)
3282 {
3283 	const struct firmware *phyf;
3284 	int ret;
3285 	struct info_10gbt_phy_fw *phy_info;
3286 
3287 	/* Use the device ID to determine which PHY file to flash.
3288 	 */
3289 	phy_info = find_phy_info(adap->pdev->device);
3290 	if (!phy_info) {
3291 		dev_warn(adap->pdev_dev,
3292 			 "No PHY Firmware file found for this PHY\n");
3293 		return -EOPNOTSUPP;
3294 	}
3295 
3296 	/* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3297 	 * use that. The adapter firmware provides us with a memory buffer
3298 	 * where we can load a PHY firmware file from the host if we want to
3299 	 * override the PHY firmware File in flash.
3300 	 */
3301 	ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3302 				      adap->pdev_dev);
3303 	if (ret < 0) {
3304 		/* For adapters without FLASH attached to PHY for their
3305 		 * firmware, it's obviously a fatal error if we can't get the
3306 		 * firmware to the adapter.  For adapters with PHY firmware
3307 		 * FLASH storage, it's worth a warning if we can't find the
3308 		 * PHY Firmware but we'll neuter the error ...
3309 		 */
3310 		dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3311 			"/lib/firmware/%s, error %d\n",
3312 			phy_info->phy_fw_file, -ret);
3313 		if (phy_info->phy_flash) {
3314 			int cur_phy_fw_ver = 0;
3315 
3316 			t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3317 			dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3318 				 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3319 			ret = 0;
3320 		}
3321 
3322 		return ret;
3323 	}
3324 
3325 	/* Load PHY Firmware onto adapter.
3326 	 */
3327 	ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3328 			     phy_info->phy_fw_version,
3329 			     (u8 *)phyf->data, phyf->size);
3330 	if (ret < 0)
3331 		dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3332 			-ret);
3333 	else if (ret > 0) {
3334 		int new_phy_fw_ver = 0;
3335 
3336 		if (phy_info->phy_fw_version)
3337 			new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3338 								  phyf->size);
3339 		dev_info(adap->pdev_dev, "Successfully transferred PHY "
3340 			 "Firmware /lib/firmware/%s, version %#x\n",
3341 			 phy_info->phy_fw_file, new_phy_fw_ver);
3342 	}
3343 
3344 	release_firmware(phyf);
3345 
3346 	return ret;
3347 }
3348 
3349 /*
3350  * Attempt to initialize the adapter via a Firmware Configuration File.
3351  */
3352 static int adap_init0_config(struct adapter *adapter, int reset)
3353 {
3354 	struct fw_caps_config_cmd caps_cmd;
3355 	const struct firmware *cf;
3356 	unsigned long mtype = 0, maddr = 0;
3357 	u32 finiver, finicsum, cfcsum;
3358 	int ret;
3359 	int config_issued = 0;
3360 	char *fw_config_file, fw_config_file_path[256];
3361 	char *config_name = NULL;
3362 
3363 	/*
3364 	 * Reset device if necessary.
3365 	 */
3366 	if (reset) {
3367 		ret = t4_fw_reset(adapter, adapter->mbox,
3368 				  PIORSTMODE_F | PIORST_F);
3369 		if (ret < 0)
3370 			goto bye;
3371 	}
3372 
3373 	/* If this is a 10Gb/s-BT adapter make sure the chip-external
3374 	 * 10Gb/s-BT PHYs have up-to-date firmware.  Note that this step needs
3375 	 * to be performed after any global adapter RESET above since some
3376 	 * PHYs only have local RAM copies of the PHY firmware.
3377 	 */
3378 	if (is_10gbt_device(adapter->pdev->device)) {
3379 		ret = adap_init0_phy(adapter);
3380 		if (ret < 0)
3381 			goto bye;
3382 	}
3383 	/*
3384 	 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3385 	 * then use that.  Otherwise, use the configuration file stored
3386 	 * in the adapter flash ...
3387 	 */
3388 	switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3389 	case CHELSIO_T4:
3390 		fw_config_file = FW4_CFNAME;
3391 		break;
3392 	case CHELSIO_T5:
3393 		fw_config_file = FW5_CFNAME;
3394 		break;
3395 	case CHELSIO_T6:
3396 		fw_config_file = FW6_CFNAME;
3397 		break;
3398 	default:
3399 		dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3400 		       adapter->pdev->device);
3401 		ret = -EINVAL;
3402 		goto bye;
3403 	}
3404 
3405 	ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3406 	if (ret < 0) {
3407 		config_name = "On FLASH";
3408 		mtype = FW_MEMTYPE_CF_FLASH;
3409 		maddr = t4_flash_cfg_addr(adapter);
3410 	} else {
3411 		u32 params[7], val[7];
3412 
3413 		sprintf(fw_config_file_path,
3414 			"/lib/firmware/%s", fw_config_file);
3415 		config_name = fw_config_file_path;
3416 
3417 		if (cf->size >= FLASH_CFG_MAX_SIZE)
3418 			ret = -ENOMEM;
3419 		else {
3420 			params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3421 			     FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3422 			ret = t4_query_params(adapter, adapter->mbox,
3423 					      adapter->pf, 0, 1, params, val);
3424 			if (ret == 0) {
3425 				/*
3426 				 * For t4_memory_rw() below addresses and
3427 				 * sizes have to be in terms of multiples of 4
3428 				 * bytes.  So, if the Configuration File isn't
3429 				 * a multiple of 4 bytes in length we'll have
3430 				 * to write that out separately since we can't
3431 				 * guarantee that the bytes following the
3432 				 * residual byte in the buffer returned by
3433 				 * request_firmware() are zeroed out ...
3434 				 */
3435 				size_t resid = cf->size & 0x3;
3436 				size_t size = cf->size & ~0x3;
3437 				__be32 *data = (__be32 *)cf->data;
3438 
3439 				mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3440 				maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3441 
3442 				spin_lock(&adapter->win0_lock);
3443 				ret = t4_memory_rw(adapter, 0, mtype, maddr,
3444 						   size, data, T4_MEMORY_WRITE);
3445 				if (ret == 0 && resid != 0) {
3446 					union {
3447 						__be32 word;
3448 						char buf[4];
3449 					} last;
3450 					int i;
3451 
3452 					last.word = data[size >> 2];
3453 					for (i = resid; i < 4; i++)
3454 						last.buf[i] = 0;
3455 					ret = t4_memory_rw(adapter, 0, mtype,
3456 							   maddr + size,
3457 							   4, &last.word,
3458 							   T4_MEMORY_WRITE);
3459 				}
3460 				spin_unlock(&adapter->win0_lock);
3461 			}
3462 		}
3463 
3464 		release_firmware(cf);
3465 		if (ret)
3466 			goto bye;
3467 	}
3468 
3469 	/*
3470 	 * Issue a Capability Configuration command to the firmware to get it
3471 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
3472 	 * because we want the ability to modify various features after we've
3473 	 * processed the configuration file ...
3474 	 */
3475 	memset(&caps_cmd, 0, sizeof(caps_cmd));
3476 	caps_cmd.op_to_write =
3477 		htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3478 		      FW_CMD_REQUEST_F |
3479 		      FW_CMD_READ_F);
3480 	caps_cmd.cfvalid_to_len16 =
3481 		htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3482 		      FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3483 		      FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3484 		      FW_LEN16(caps_cmd));
3485 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3486 			 &caps_cmd);
3487 
3488 	/* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3489 	 * Configuration File in FLASH), our last gasp effort is to use the
3490 	 * Firmware Configuration File which is embedded in the firmware.  A
3491 	 * very few early versions of the firmware didn't have one embedded
3492 	 * but we can ignore those.
3493 	 */
3494 	if (ret == -ENOENT) {
3495 		memset(&caps_cmd, 0, sizeof(caps_cmd));
3496 		caps_cmd.op_to_write =
3497 			htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3498 					FW_CMD_REQUEST_F |
3499 					FW_CMD_READ_F);
3500 		caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3501 		ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3502 				sizeof(caps_cmd), &caps_cmd);
3503 		config_name = "Firmware Default";
3504 	}
3505 
3506 	config_issued = 1;
3507 	if (ret < 0)
3508 		goto bye;
3509 
3510 	finiver = ntohl(caps_cmd.finiver);
3511 	finicsum = ntohl(caps_cmd.finicsum);
3512 	cfcsum = ntohl(caps_cmd.cfcsum);
3513 	if (finicsum != cfcsum)
3514 		dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3515 			 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3516 			 finicsum, cfcsum);
3517 
3518 	/*
3519 	 * And now tell the firmware to use the configuration we just loaded.
3520 	 */
3521 	caps_cmd.op_to_write =
3522 		htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3523 		      FW_CMD_REQUEST_F |
3524 		      FW_CMD_WRITE_F);
3525 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3526 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3527 			 NULL);
3528 	if (ret < 0)
3529 		goto bye;
3530 
3531 	/*
3532 	 * Tweak configuration based on system architecture, module
3533 	 * parameters, etc.
3534 	 */
3535 	ret = adap_init0_tweaks(adapter);
3536 	if (ret < 0)
3537 		goto bye;
3538 
3539 	/*
3540 	 * And finally tell the firmware to initialize itself using the
3541 	 * parameters from the Configuration File.
3542 	 */
3543 	ret = t4_fw_initialize(adapter, adapter->mbox);
3544 	if (ret < 0)
3545 		goto bye;
3546 
3547 	/* Emit Firmware Configuration File information and return
3548 	 * successfully.
3549 	 */
3550 	dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3551 		 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3552 		 config_name, finiver, cfcsum);
3553 	return 0;
3554 
3555 	/*
3556 	 * Something bad happened.  Return the error ...  (If the "error"
3557 	 * is that there's no Configuration File on the adapter we don't
3558 	 * want to issue a warning since this is fairly common.)
3559 	 */
3560 bye:
3561 	if (config_issued && ret != -ENOENT)
3562 		dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3563 			 config_name, -ret);
3564 	return ret;
3565 }
3566 
3567 static struct fw_info fw_info_array[] = {
3568 	{
3569 		.chip = CHELSIO_T4,
3570 		.fs_name = FW4_CFNAME,
3571 		.fw_mod_name = FW4_FNAME,
3572 		.fw_hdr = {
3573 			.chip = FW_HDR_CHIP_T4,
3574 			.fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3575 			.intfver_nic = FW_INTFVER(T4, NIC),
3576 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3577 			.intfver_ri = FW_INTFVER(T4, RI),
3578 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3579 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3580 		},
3581 	}, {
3582 		.chip = CHELSIO_T5,
3583 		.fs_name = FW5_CFNAME,
3584 		.fw_mod_name = FW5_FNAME,
3585 		.fw_hdr = {
3586 			.chip = FW_HDR_CHIP_T5,
3587 			.fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3588 			.intfver_nic = FW_INTFVER(T5, NIC),
3589 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3590 			.intfver_ri = FW_INTFVER(T5, RI),
3591 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3592 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3593 		},
3594 	}, {
3595 		.chip = CHELSIO_T6,
3596 		.fs_name = FW6_CFNAME,
3597 		.fw_mod_name = FW6_FNAME,
3598 		.fw_hdr = {
3599 			.chip = FW_HDR_CHIP_T6,
3600 			.fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3601 			.intfver_nic = FW_INTFVER(T6, NIC),
3602 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3603 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3604 			.intfver_ri = FW_INTFVER(T6, RI),
3605 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3606 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3607 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3608 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3609 		},
3610 	}
3611 
3612 };
3613 
3614 static struct fw_info *find_fw_info(int chip)
3615 {
3616 	int i;
3617 
3618 	for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3619 		if (fw_info_array[i].chip == chip)
3620 			return &fw_info_array[i];
3621 	}
3622 	return NULL;
3623 }
3624 
3625 /*
3626  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3627  */
3628 static int adap_init0(struct adapter *adap)
3629 {
3630 	int ret;
3631 	u32 v, port_vec;
3632 	enum dev_state state;
3633 	u32 params[7], val[7];
3634 	struct fw_caps_config_cmd caps_cmd;
3635 	int reset = 1;
3636 
3637 	/* Grab Firmware Device Log parameters as early as possible so we have
3638 	 * access to it for debugging, etc.
3639 	 */
3640 	ret = t4_init_devlog_params(adap);
3641 	if (ret < 0)
3642 		return ret;
3643 
3644 	/* Contact FW, advertising Master capability */
3645 	ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
3646 	if (ret < 0) {
3647 		dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3648 			ret);
3649 		return ret;
3650 	}
3651 	if (ret == adap->mbox)
3652 		adap->flags |= MASTER_PF;
3653 
3654 	/*
3655 	 * If we're the Master PF Driver and the device is uninitialized,
3656 	 * then let's consider upgrading the firmware ...  (We always want
3657 	 * to check the firmware version number in order to A. get it for
3658 	 * later reporting and B. to warn if the currently loaded firmware
3659 	 * is excessively mismatched relative to the driver.)
3660 	 */
3661 	t4_get_fw_version(adap, &adap->params.fw_vers);
3662 	t4_get_tp_version(adap, &adap->params.tp_vers);
3663 	if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
3664 		struct fw_info *fw_info;
3665 		struct fw_hdr *card_fw;
3666 		const struct firmware *fw;
3667 		const u8 *fw_data = NULL;
3668 		unsigned int fw_size = 0;
3669 
3670 		/* This is the firmware whose headers the driver was compiled
3671 		 * against
3672 		 */
3673 		fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3674 		if (fw_info == NULL) {
3675 			dev_err(adap->pdev_dev,
3676 				"unable to get firmware info for chip %d.\n",
3677 				CHELSIO_CHIP_VERSION(adap->params.chip));
3678 			return -EINVAL;
3679 		}
3680 
3681 		/* allocate memory to read the header of the firmware on the
3682 		 * card
3683 		 */
3684 		card_fw = t4_alloc_mem(sizeof(*card_fw));
3685 
3686 		/* Get FW from from /lib/firmware/ */
3687 		ret = request_firmware(&fw, fw_info->fw_mod_name,
3688 				       adap->pdev_dev);
3689 		if (ret < 0) {
3690 			dev_err(adap->pdev_dev,
3691 				"unable to load firmware image %s, error %d\n",
3692 				fw_info->fw_mod_name, ret);
3693 		} else {
3694 			fw_data = fw->data;
3695 			fw_size = fw->size;
3696 		}
3697 
3698 		/* upgrade FW logic */
3699 		ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3700 				 state, &reset);
3701 
3702 		/* Cleaning up */
3703 		release_firmware(fw);
3704 		t4_free_mem(card_fw);
3705 
3706 		if (ret < 0)
3707 			goto bye;
3708 	}
3709 
3710 	/*
3711 	 * Grab VPD parameters.  This should be done after we establish a
3712 	 * connection to the firmware since some of the VPD parameters
3713 	 * (notably the Core Clock frequency) are retrieved via requests to
3714 	 * the firmware.  On the other hand, we need these fairly early on
3715 	 * so we do this right after getting ahold of the firmware.
3716 	 */
3717 	ret = t4_get_vpd_params(adap, &adap->params.vpd);
3718 	if (ret < 0)
3719 		goto bye;
3720 
3721 	/*
3722 	 * Find out what ports are available to us.  Note that we need to do
3723 	 * this before calling adap_init0_no_config() since it needs nports
3724 	 * and portvec ...
3725 	 */
3726 	v =
3727 	    FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3728 	    FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
3729 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
3730 	if (ret < 0)
3731 		goto bye;
3732 
3733 	adap->params.nports = hweight32(port_vec);
3734 	adap->params.portvec = port_vec;
3735 
3736 	/* If the firmware is initialized already, emit a simply note to that
3737 	 * effect. Otherwise, it's time to try initializing the adapter.
3738 	 */
3739 	if (state == DEV_STATE_INIT) {
3740 		dev_info(adap->pdev_dev, "Coming up as %s: "\
3741 			 "Adapter already initialized\n",
3742 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
3743 	} else {
3744 		dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3745 			 "Initializing adapter\n");
3746 
3747 		/* Find out whether we're dealing with a version of the
3748 		 * firmware which has configuration file support.
3749 		 */
3750 		params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3751 			     FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3752 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
3753 				      params, val);
3754 
3755 		/* If the firmware doesn't support Configuration Files,
3756 		 * return an error.
3757 		 */
3758 		if (ret < 0) {
3759 			dev_err(adap->pdev_dev, "firmware doesn't support "
3760 				"Firmware Configuration Files\n");
3761 			goto bye;
3762 		}
3763 
3764 		/* The firmware provides us with a memory buffer where we can
3765 		 * load a Configuration File from the host if we want to
3766 		 * override the Configuration File in flash.
3767 		 */
3768 		ret = adap_init0_config(adap, reset);
3769 		if (ret == -ENOENT) {
3770 			dev_err(adap->pdev_dev, "no Configuration File "
3771 				"present on adapter.\n");
3772 			goto bye;
3773 		}
3774 		if (ret < 0) {
3775 			dev_err(adap->pdev_dev, "could not initialize "
3776 				"adapter, error %d\n", -ret);
3777 			goto bye;
3778 		}
3779 	}
3780 
3781 	/* Give the SGE code a chance to pull in anything that it needs ...
3782 	 * Note that this must be called after we retrieve our VPD parameters
3783 	 * in order to know how to convert core ticks to seconds, etc.
3784 	 */
3785 	ret = t4_sge_init(adap);
3786 	if (ret < 0)
3787 		goto bye;
3788 
3789 	if (is_bypass_device(adap->pdev->device))
3790 		adap->params.bypass = 1;
3791 
3792 	/*
3793 	 * Grab some of our basic fundamental operating parameters.
3794 	 */
3795 #define FW_PARAM_DEV(param) \
3796 	(FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
3797 	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
3798 
3799 #define FW_PARAM_PFVF(param) \
3800 	FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
3801 	FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)|  \
3802 	FW_PARAMS_PARAM_Y_V(0) | \
3803 	FW_PARAMS_PARAM_Z_V(0)
3804 
3805 	params[0] = FW_PARAM_PFVF(EQ_START);
3806 	params[1] = FW_PARAM_PFVF(L2T_START);
3807 	params[2] = FW_PARAM_PFVF(L2T_END);
3808 	params[3] = FW_PARAM_PFVF(FILTER_START);
3809 	params[4] = FW_PARAM_PFVF(FILTER_END);
3810 	params[5] = FW_PARAM_PFVF(IQFLINT_START);
3811 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
3812 	if (ret < 0)
3813 		goto bye;
3814 	adap->sge.egr_start = val[0];
3815 	adap->l2t_start = val[1];
3816 	adap->l2t_end = val[2];
3817 	adap->tids.ftid_base = val[3];
3818 	adap->tids.nftids = val[4] - val[3] + 1;
3819 	adap->sge.ingr_start = val[5];
3820 
3821 	/* qids (ingress/egress) returned from firmware can be anywhere
3822 	 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
3823 	 * Hence driver needs to allocate memory for this range to
3824 	 * store the queue info. Get the highest IQFLINT/EQ index returned
3825 	 * in FW_EQ_*_CMD.alloc command.
3826 	 */
3827 	params[0] = FW_PARAM_PFVF(EQ_END);
3828 	params[1] = FW_PARAM_PFVF(IQFLINT_END);
3829 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3830 	if (ret < 0)
3831 		goto bye;
3832 	adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
3833 	adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
3834 
3835 	adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
3836 				    sizeof(*adap->sge.egr_map), GFP_KERNEL);
3837 	if (!adap->sge.egr_map) {
3838 		ret = -ENOMEM;
3839 		goto bye;
3840 	}
3841 
3842 	adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
3843 				     sizeof(*adap->sge.ingr_map), GFP_KERNEL);
3844 	if (!adap->sge.ingr_map) {
3845 		ret = -ENOMEM;
3846 		goto bye;
3847 	}
3848 
3849 	/* Allocate the memory for the vaious egress queue bitmaps
3850 	 * ie starving_fl, txq_maperr and blocked_fl.
3851 	 */
3852 	adap->sge.starving_fl =	kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3853 					sizeof(long), GFP_KERNEL);
3854 	if (!adap->sge.starving_fl) {
3855 		ret = -ENOMEM;
3856 		goto bye;
3857 	}
3858 
3859 	adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3860 				       sizeof(long), GFP_KERNEL);
3861 	if (!adap->sge.txq_maperr) {
3862 		ret = -ENOMEM;
3863 		goto bye;
3864 	}
3865 
3866 #ifdef CONFIG_DEBUG_FS
3867 	adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3868 				       sizeof(long), GFP_KERNEL);
3869 	if (!adap->sge.blocked_fl) {
3870 		ret = -ENOMEM;
3871 		goto bye;
3872 	}
3873 #endif
3874 
3875 	params[0] = FW_PARAM_PFVF(CLIP_START);
3876 	params[1] = FW_PARAM_PFVF(CLIP_END);
3877 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3878 	if (ret < 0)
3879 		goto bye;
3880 	adap->clipt_start = val[0];
3881 	adap->clipt_end = val[1];
3882 
3883 	/* query params related to active filter region */
3884 	params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
3885 	params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
3886 	ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3887 	/* If Active filter size is set we enable establishing
3888 	 * offload connection through firmware work request
3889 	 */
3890 	if ((val[0] != val[1]) && (ret >= 0)) {
3891 		adap->flags |= FW_OFLD_CONN;
3892 		adap->tids.aftid_base = val[0];
3893 		adap->tids.aftid_end = val[1];
3894 	}
3895 
3896 	/* If we're running on newer firmware, let it know that we're
3897 	 * prepared to deal with encapsulated CPL messages.  Older
3898 	 * firmware won't understand this and we'll just get
3899 	 * unencapsulated messages ...
3900 	 */
3901 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3902 	val[0] = 1;
3903 	(void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
3904 
3905 	/*
3906 	 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
3907 	 * capability.  Earlier versions of the firmware didn't have the
3908 	 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
3909 	 * permission to use ULPTX MEMWRITE DSGL.
3910 	 */
3911 	if (is_t4(adap->params.chip)) {
3912 		adap->params.ulptx_memwrite_dsgl = false;
3913 	} else {
3914 		params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
3915 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3916 				      1, params, val);
3917 		adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
3918 	}
3919 
3920 	/*
3921 	 * Get device capabilities so we can determine what resources we need
3922 	 * to manage.
3923 	 */
3924 	memset(&caps_cmd, 0, sizeof(caps_cmd));
3925 	caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3926 				     FW_CMD_REQUEST_F | FW_CMD_READ_F);
3927 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3928 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
3929 			 &caps_cmd);
3930 	if (ret < 0)
3931 		goto bye;
3932 
3933 	if (caps_cmd.ofldcaps) {
3934 		/* query offload-related parameters */
3935 		params[0] = FW_PARAM_DEV(NTID);
3936 		params[1] = FW_PARAM_PFVF(SERVER_START);
3937 		params[2] = FW_PARAM_PFVF(SERVER_END);
3938 		params[3] = FW_PARAM_PFVF(TDDP_START);
3939 		params[4] = FW_PARAM_PFVF(TDDP_END);
3940 		params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3941 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3942 				      params, val);
3943 		if (ret < 0)
3944 			goto bye;
3945 		adap->tids.ntids = val[0];
3946 		adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3947 		adap->tids.stid_base = val[1];
3948 		adap->tids.nstids = val[2] - val[1] + 1;
3949 		/*
3950 		 * Setup server filter region. Divide the available filter
3951 		 * region into two parts. Regular filters get 1/3rd and server
3952 		 * filters get 2/3rd part. This is only enabled if workarond
3953 		 * path is enabled.
3954 		 * 1. For regular filters.
3955 		 * 2. Server filter: This are special filters which are used
3956 		 * to redirect SYN packets to offload queue.
3957 		 */
3958 		if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
3959 			adap->tids.sftid_base = adap->tids.ftid_base +
3960 					DIV_ROUND_UP(adap->tids.nftids, 3);
3961 			adap->tids.nsftids = adap->tids.nftids -
3962 					 DIV_ROUND_UP(adap->tids.nftids, 3);
3963 			adap->tids.nftids = adap->tids.sftid_base -
3964 						adap->tids.ftid_base;
3965 		}
3966 		adap->vres.ddp.start = val[3];
3967 		adap->vres.ddp.size = val[4] - val[3] + 1;
3968 		adap->params.ofldq_wr_cred = val[5];
3969 
3970 		adap->params.offload = 1;
3971 	}
3972 	if (caps_cmd.rdmacaps) {
3973 		params[0] = FW_PARAM_PFVF(STAG_START);
3974 		params[1] = FW_PARAM_PFVF(STAG_END);
3975 		params[2] = FW_PARAM_PFVF(RQ_START);
3976 		params[3] = FW_PARAM_PFVF(RQ_END);
3977 		params[4] = FW_PARAM_PFVF(PBL_START);
3978 		params[5] = FW_PARAM_PFVF(PBL_END);
3979 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3980 				      params, val);
3981 		if (ret < 0)
3982 			goto bye;
3983 		adap->vres.stag.start = val[0];
3984 		adap->vres.stag.size = val[1] - val[0] + 1;
3985 		adap->vres.rq.start = val[2];
3986 		adap->vres.rq.size = val[3] - val[2] + 1;
3987 		adap->vres.pbl.start = val[4];
3988 		adap->vres.pbl.size = val[5] - val[4] + 1;
3989 
3990 		params[0] = FW_PARAM_PFVF(SQRQ_START);
3991 		params[1] = FW_PARAM_PFVF(SQRQ_END);
3992 		params[2] = FW_PARAM_PFVF(CQ_START);
3993 		params[3] = FW_PARAM_PFVF(CQ_END);
3994 		params[4] = FW_PARAM_PFVF(OCQ_START);
3995 		params[5] = FW_PARAM_PFVF(OCQ_END);
3996 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
3997 				      val);
3998 		if (ret < 0)
3999 			goto bye;
4000 		adap->vres.qp.start = val[0];
4001 		adap->vres.qp.size = val[1] - val[0] + 1;
4002 		adap->vres.cq.start = val[2];
4003 		adap->vres.cq.size = val[3] - val[2] + 1;
4004 		adap->vres.ocq.start = val[4];
4005 		adap->vres.ocq.size = val[5] - val[4] + 1;
4006 
4007 		params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
4008 		params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4009 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
4010 				      val);
4011 		if (ret < 0) {
4012 			adap->params.max_ordird_qp = 8;
4013 			adap->params.max_ird_adapter = 32 * adap->tids.ntids;
4014 			ret = 0;
4015 		} else {
4016 			adap->params.max_ordird_qp = val[0];
4017 			adap->params.max_ird_adapter = val[1];
4018 		}
4019 		dev_info(adap->pdev_dev,
4020 			 "max_ordird_qp %d max_ird_adapter %d\n",
4021 			 adap->params.max_ordird_qp,
4022 			 adap->params.max_ird_adapter);
4023 	}
4024 	if (caps_cmd.iscsicaps) {
4025 		params[0] = FW_PARAM_PFVF(ISCSI_START);
4026 		params[1] = FW_PARAM_PFVF(ISCSI_END);
4027 		ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4028 				      params, val);
4029 		if (ret < 0)
4030 			goto bye;
4031 		adap->vres.iscsi.start = val[0];
4032 		adap->vres.iscsi.size = val[1] - val[0] + 1;
4033 	}
4034 #undef FW_PARAM_PFVF
4035 #undef FW_PARAM_DEV
4036 
4037 	/* The MTU/MSS Table is initialized by now, so load their values.  If
4038 	 * we're initializing the adapter, then we'll make any modifications
4039 	 * we want to the MTU/MSS Table and also initialize the congestion
4040 	 * parameters.
4041 	 */
4042 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4043 	if (state != DEV_STATE_INIT) {
4044 		int i;
4045 
4046 		/* The default MTU Table contains values 1492 and 1500.
4047 		 * However, for TCP, it's better to have two values which are
4048 		 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
4049 		 * This allows us to have a TCP Data Payload which is a
4050 		 * multiple of 8 regardless of what combination of TCP Options
4051 		 * are in use (always a multiple of 4 bytes) which is
4052 		 * important for performance reasons.  For instance, if no
4053 		 * options are in use, then we have a 20-byte IP header and a
4054 		 * 20-byte TCP header.  In this case, a 1500-byte MSS would
4055 		 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
4056 		 * which is not a multiple of 8.  So using an MSS of 1488 in
4057 		 * this case results in a TCP Data Payload of 1448 bytes which
4058 		 * is a multiple of 8.  On the other hand, if 12-byte TCP Time
4059 		 * Stamps have been negotiated, then an MTU of 1500 bytes
4060 		 * results in a TCP Data Payload of 1448 bytes which, as
4061 		 * above, is a multiple of 8 bytes ...
4062 		 */
4063 		for (i = 0; i < NMTUS; i++)
4064 			if (adap->params.mtus[i] == 1492) {
4065 				adap->params.mtus[i] = 1488;
4066 				break;
4067 			}
4068 
4069 		t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4070 			     adap->params.b_wnd);
4071 	}
4072 	t4_init_sge_params(adap);
4073 	adap->flags |= FW_OK;
4074 	t4_init_tp_params(adap);
4075 	return 0;
4076 
4077 	/*
4078 	 * Something bad happened.  If a command timed out or failed with EIO
4079 	 * FW does not operate within its spec or something catastrophic
4080 	 * happened to HW/FW, stop issuing commands.
4081 	 */
4082 bye:
4083 	kfree(adap->sge.egr_map);
4084 	kfree(adap->sge.ingr_map);
4085 	kfree(adap->sge.starving_fl);
4086 	kfree(adap->sge.txq_maperr);
4087 #ifdef CONFIG_DEBUG_FS
4088 	kfree(adap->sge.blocked_fl);
4089 #endif
4090 	if (ret != -ETIMEDOUT && ret != -EIO)
4091 		t4_fw_bye(adap, adap->mbox);
4092 	return ret;
4093 }
4094 
4095 /* EEH callbacks */
4096 
4097 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4098 					 pci_channel_state_t state)
4099 {
4100 	int i;
4101 	struct adapter *adap = pci_get_drvdata(pdev);
4102 
4103 	if (!adap)
4104 		goto out;
4105 
4106 	rtnl_lock();
4107 	adap->flags &= ~FW_OK;
4108 	notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4109 	spin_lock(&adap->stats_lock);
4110 	for_each_port(adap, i) {
4111 		struct net_device *dev = adap->port[i];
4112 
4113 		netif_device_detach(dev);
4114 		netif_carrier_off(dev);
4115 	}
4116 	spin_unlock(&adap->stats_lock);
4117 	disable_interrupts(adap);
4118 	if (adap->flags & FULL_INIT_DONE)
4119 		cxgb_down(adap);
4120 	rtnl_unlock();
4121 	if ((adap->flags & DEV_ENABLED)) {
4122 		pci_disable_device(pdev);
4123 		adap->flags &= ~DEV_ENABLED;
4124 	}
4125 out:	return state == pci_channel_io_perm_failure ?
4126 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4127 }
4128 
4129 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4130 {
4131 	int i, ret;
4132 	struct fw_caps_config_cmd c;
4133 	struct adapter *adap = pci_get_drvdata(pdev);
4134 
4135 	if (!adap) {
4136 		pci_restore_state(pdev);
4137 		pci_save_state(pdev);
4138 		return PCI_ERS_RESULT_RECOVERED;
4139 	}
4140 
4141 	if (!(adap->flags & DEV_ENABLED)) {
4142 		if (pci_enable_device(pdev)) {
4143 			dev_err(&pdev->dev, "Cannot reenable PCI "
4144 					    "device after reset\n");
4145 			return PCI_ERS_RESULT_DISCONNECT;
4146 		}
4147 		adap->flags |= DEV_ENABLED;
4148 	}
4149 
4150 	pci_set_master(pdev);
4151 	pci_restore_state(pdev);
4152 	pci_save_state(pdev);
4153 	pci_cleanup_aer_uncorrect_error_status(pdev);
4154 
4155 	if (t4_wait_dev_ready(adap->regs) < 0)
4156 		return PCI_ERS_RESULT_DISCONNECT;
4157 	if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
4158 		return PCI_ERS_RESULT_DISCONNECT;
4159 	adap->flags |= FW_OK;
4160 	if (adap_init1(adap, &c))
4161 		return PCI_ERS_RESULT_DISCONNECT;
4162 
4163 	for_each_port(adap, i) {
4164 		struct port_info *p = adap2pinfo(adap, i);
4165 
4166 		ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
4167 				  NULL, NULL);
4168 		if (ret < 0)
4169 			return PCI_ERS_RESULT_DISCONNECT;
4170 		p->viid = ret;
4171 		p->xact_addr_filt = -1;
4172 	}
4173 
4174 	t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4175 		     adap->params.b_wnd);
4176 	setup_memwin(adap);
4177 	if (cxgb_up(adap))
4178 		return PCI_ERS_RESULT_DISCONNECT;
4179 	return PCI_ERS_RESULT_RECOVERED;
4180 }
4181 
4182 static void eeh_resume(struct pci_dev *pdev)
4183 {
4184 	int i;
4185 	struct adapter *adap = pci_get_drvdata(pdev);
4186 
4187 	if (!adap)
4188 		return;
4189 
4190 	rtnl_lock();
4191 	for_each_port(adap, i) {
4192 		struct net_device *dev = adap->port[i];
4193 
4194 		if (netif_running(dev)) {
4195 			link_start(dev);
4196 			cxgb_set_rxmode(dev);
4197 		}
4198 		netif_device_attach(dev);
4199 	}
4200 	rtnl_unlock();
4201 }
4202 
4203 static const struct pci_error_handlers cxgb4_eeh = {
4204 	.error_detected = eeh_err_detected,
4205 	.slot_reset     = eeh_slot_reset,
4206 	.resume         = eeh_resume,
4207 };
4208 
4209 static inline bool is_x_10g_port(const struct link_config *lc)
4210 {
4211 	return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0 ||
4212 	       (lc->supported & FW_PORT_CAP_SPEED_40G) != 0;
4213 }
4214 
4215 static inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
4216 			     unsigned int us, unsigned int cnt,
4217 			     unsigned int size, unsigned int iqe_size)
4218 {
4219 	q->adap = adap;
4220 	cxgb4_set_rspq_intr_params(q, us, cnt);
4221 	q->iqe_len = iqe_size;
4222 	q->size = size;
4223 }
4224 
4225 /*
4226  * Perform default configuration of DMA queues depending on the number and type
4227  * of ports we found and the number of available CPUs.  Most settings can be
4228  * modified by the admin prior to actual use.
4229  */
4230 static void cfg_queues(struct adapter *adap)
4231 {
4232 	struct sge *s = &adap->sge;
4233 	int i, n10g = 0, qidx = 0;
4234 #ifndef CONFIG_CHELSIO_T4_DCB
4235 	int q10g = 0;
4236 #endif
4237 	int ciq_size;
4238 
4239 	for_each_port(adap, i)
4240 		n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4241 #ifdef CONFIG_CHELSIO_T4_DCB
4242 	/* For Data Center Bridging support we need to be able to support up
4243 	 * to 8 Traffic Priorities; each of which will be assigned to its
4244 	 * own TX Queue in order to prevent Head-Of-Line Blocking.
4245 	 */
4246 	if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4247 		dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4248 			MAX_ETH_QSETS, adap->params.nports * 8);
4249 		BUG_ON(1);
4250 	}
4251 
4252 	for_each_port(adap, i) {
4253 		struct port_info *pi = adap2pinfo(adap, i);
4254 
4255 		pi->first_qset = qidx;
4256 		pi->nqsets = 8;
4257 		qidx += pi->nqsets;
4258 	}
4259 #else /* !CONFIG_CHELSIO_T4_DCB */
4260 	/*
4261 	 * We default to 1 queue per non-10G port and up to # of cores queues
4262 	 * per 10G port.
4263 	 */
4264 	if (n10g)
4265 		q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4266 	if (q10g > netif_get_num_default_rss_queues())
4267 		q10g = netif_get_num_default_rss_queues();
4268 
4269 	for_each_port(adap, i) {
4270 		struct port_info *pi = adap2pinfo(adap, i);
4271 
4272 		pi->first_qset = qidx;
4273 		pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4274 		qidx += pi->nqsets;
4275 	}
4276 #endif /* !CONFIG_CHELSIO_T4_DCB */
4277 
4278 	s->ethqsets = qidx;
4279 	s->max_ethqsets = qidx;   /* MSI-X may lower it later */
4280 
4281 	if (is_offload(adap)) {
4282 		/*
4283 		 * For offload we use 1 queue/channel if all ports are up to 1G,
4284 		 * otherwise we divide all available queues amongst the channels
4285 		 * capped by the number of available cores.
4286 		 */
4287 		if (n10g) {
4288 			i = min_t(int, ARRAY_SIZE(s->ofldrxq),
4289 				  num_online_cpus());
4290 			s->ofldqsets = roundup(i, adap->params.nports);
4291 		} else
4292 			s->ofldqsets = adap->params.nports;
4293 		/* For RDMA one Rx queue per channel suffices */
4294 		s->rdmaqs = adap->params.nports;
4295 		/* Try and allow at least 1 CIQ per cpu rounding down
4296 		 * to the number of ports, with a minimum of 1 per port.
4297 		 * A 2 port card in a 6 cpu system: 6 CIQs, 3 / port.
4298 		 * A 4 port card in a 6 cpu system: 4 CIQs, 1 / port.
4299 		 * A 4 port card in a 2 cpu system: 4 CIQs, 1 / port.
4300 		 */
4301 		s->rdmaciqs = min_t(int, MAX_RDMA_CIQS, num_online_cpus());
4302 		s->rdmaciqs = (s->rdmaciqs / adap->params.nports) *
4303 				adap->params.nports;
4304 		s->rdmaciqs = max_t(int, s->rdmaciqs, adap->params.nports);
4305 	}
4306 
4307 	for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4308 		struct sge_eth_rxq *r = &s->ethrxq[i];
4309 
4310 		init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4311 		r->fl.size = 72;
4312 	}
4313 
4314 	for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4315 		s->ethtxq[i].q.size = 1024;
4316 
4317 	for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4318 		s->ctrlq[i].q.size = 512;
4319 
4320 	for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
4321 		s->ofldtxq[i].q.size = 1024;
4322 
4323 	for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
4324 		struct sge_ofld_rxq *r = &s->ofldrxq[i];
4325 
4326 		init_rspq(adap, &r->rspq, 5, 1, 1024, 64);
4327 		r->rspq.uld = CXGB4_ULD_ISCSI;
4328 		r->fl.size = 72;
4329 	}
4330 
4331 	for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
4332 		struct sge_ofld_rxq *r = &s->rdmarxq[i];
4333 
4334 		init_rspq(adap, &r->rspq, 5, 1, 511, 64);
4335 		r->rspq.uld = CXGB4_ULD_RDMA;
4336 		r->fl.size = 72;
4337 	}
4338 
4339 	ciq_size = 64 + adap->vres.cq.size + adap->tids.nftids;
4340 	if (ciq_size > SGE_MAX_IQ_SIZE) {
4341 		CH_WARN(adap, "CIQ size too small for available IQs\n");
4342 		ciq_size = SGE_MAX_IQ_SIZE;
4343 	}
4344 
4345 	for (i = 0; i < ARRAY_SIZE(s->rdmaciq); i++) {
4346 		struct sge_ofld_rxq *r = &s->rdmaciq[i];
4347 
4348 		init_rspq(adap, &r->rspq, 5, 1, ciq_size, 64);
4349 		r->rspq.uld = CXGB4_ULD_RDMA;
4350 	}
4351 
4352 	init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4353 	init_rspq(adap, &s->intrq, 0, 1, 2 * MAX_INGQ, 64);
4354 }
4355 
4356 /*
4357  * Reduce the number of Ethernet queues across all ports to at most n.
4358  * n provides at least one queue per port.
4359  */
4360 static void reduce_ethqs(struct adapter *adap, int n)
4361 {
4362 	int i;
4363 	struct port_info *pi;
4364 
4365 	while (n < adap->sge.ethqsets)
4366 		for_each_port(adap, i) {
4367 			pi = adap2pinfo(adap, i);
4368 			if (pi->nqsets > 1) {
4369 				pi->nqsets--;
4370 				adap->sge.ethqsets--;
4371 				if (adap->sge.ethqsets <= n)
4372 					break;
4373 			}
4374 		}
4375 
4376 	n = 0;
4377 	for_each_port(adap, i) {
4378 		pi = adap2pinfo(adap, i);
4379 		pi->first_qset = n;
4380 		n += pi->nqsets;
4381 	}
4382 }
4383 
4384 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4385 #define EXTRA_VECS 2
4386 
4387 static int enable_msix(struct adapter *adap)
4388 {
4389 	int ofld_need = 0;
4390 	int i, want, need, allocated;
4391 	struct sge *s = &adap->sge;
4392 	unsigned int nchan = adap->params.nports;
4393 	struct msix_entry *entries;
4394 
4395 	entries = kmalloc(sizeof(*entries) * (MAX_INGQ + 1),
4396 			  GFP_KERNEL);
4397 	if (!entries)
4398 		return -ENOMEM;
4399 
4400 	for (i = 0; i < MAX_INGQ + 1; ++i)
4401 		entries[i].entry = i;
4402 
4403 	want = s->max_ethqsets + EXTRA_VECS;
4404 	if (is_offload(adap)) {
4405 		want += s->rdmaqs + s->rdmaciqs + s->ofldqsets;
4406 		/* need nchan for each possible ULD */
4407 		ofld_need = 3 * nchan;
4408 	}
4409 #ifdef CONFIG_CHELSIO_T4_DCB
4410 	/* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4411 	 * each port.
4412 	 */
4413 	need = 8 * adap->params.nports + EXTRA_VECS + ofld_need;
4414 #else
4415 	need = adap->params.nports + EXTRA_VECS + ofld_need;
4416 #endif
4417 	allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4418 	if (allocated < 0) {
4419 		dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4420 			 " not using MSI-X\n");
4421 		kfree(entries);
4422 		return allocated;
4423 	}
4424 
4425 	/* Distribute available vectors to the various queue groups.
4426 	 * Every group gets its minimum requirement and NIC gets top
4427 	 * priority for leftovers.
4428 	 */
4429 	i = allocated - EXTRA_VECS - ofld_need;
4430 	if (i < s->max_ethqsets) {
4431 		s->max_ethqsets = i;
4432 		if (i < s->ethqsets)
4433 			reduce_ethqs(adap, i);
4434 	}
4435 	if (is_offload(adap)) {
4436 		if (allocated < want) {
4437 			s->rdmaqs = nchan;
4438 			s->rdmaciqs = nchan;
4439 		}
4440 
4441 		/* leftovers go to OFLD */
4442 		i = allocated - EXTRA_VECS - s->max_ethqsets -
4443 		    s->rdmaqs - s->rdmaciqs;
4444 		s->ofldqsets = (i / nchan) * nchan;  /* round down */
4445 	}
4446 	for (i = 0; i < allocated; ++i)
4447 		adap->msix_info[i].vec = entries[i].vector;
4448 
4449 	kfree(entries);
4450 	return 0;
4451 }
4452 
4453 #undef EXTRA_VECS
4454 
4455 static int init_rss(struct adapter *adap)
4456 {
4457 	unsigned int i;
4458 	int err;
4459 
4460 	err = t4_init_rss_mode(adap, adap->mbox);
4461 	if (err)
4462 		return err;
4463 
4464 	for_each_port(adap, i) {
4465 		struct port_info *pi = adap2pinfo(adap, i);
4466 
4467 		pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4468 		if (!pi->rss)
4469 			return -ENOMEM;
4470 	}
4471 	return 0;
4472 }
4473 
4474 static void print_port_info(const struct net_device *dev)
4475 {
4476 	char buf[80];
4477 	char *bufp = buf;
4478 	const char *spd = "";
4479 	const struct port_info *pi = netdev_priv(dev);
4480 	const struct adapter *adap = pi->adapter;
4481 
4482 	if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4483 		spd = " 2.5 GT/s";
4484 	else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4485 		spd = " 5 GT/s";
4486 	else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4487 		spd = " 8 GT/s";
4488 
4489 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
4490 		bufp += sprintf(bufp, "100/");
4491 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
4492 		bufp += sprintf(bufp, "1000/");
4493 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4494 		bufp += sprintf(bufp, "10G/");
4495 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
4496 		bufp += sprintf(bufp, "40G/");
4497 	if (bufp != buf)
4498 		--bufp;
4499 	sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
4500 
4501 	netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
4502 		    adap->params.vpd.id,
4503 		    CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
4504 		    is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
4505 		    (adap->flags & USING_MSIX) ? " MSI-X" :
4506 		    (adap->flags & USING_MSI) ? " MSI" : "");
4507 	netdev_info(dev, "S/N: %s, P/N: %s\n",
4508 		    adap->params.vpd.sn, adap->params.vpd.pn);
4509 }
4510 
4511 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
4512 {
4513 	pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
4514 }
4515 
4516 /*
4517  * Free the following resources:
4518  * - memory used for tables
4519  * - MSI/MSI-X
4520  * - net devices
4521  * - resources FW is holding for us
4522  */
4523 static void free_some_resources(struct adapter *adapter)
4524 {
4525 	unsigned int i;
4526 
4527 	t4_free_mem(adapter->l2t);
4528 	t4_free_mem(adapter->tids.tid_tab);
4529 	kfree(adapter->sge.egr_map);
4530 	kfree(adapter->sge.ingr_map);
4531 	kfree(adapter->sge.starving_fl);
4532 	kfree(adapter->sge.txq_maperr);
4533 #ifdef CONFIG_DEBUG_FS
4534 	kfree(adapter->sge.blocked_fl);
4535 #endif
4536 	disable_msi(adapter);
4537 
4538 	for_each_port(adapter, i)
4539 		if (adapter->port[i]) {
4540 			struct port_info *pi = adap2pinfo(adapter, i);
4541 
4542 			if (pi->viid != 0)
4543 				t4_free_vi(adapter, adapter->mbox, adapter->pf,
4544 					   0, pi->viid);
4545 			kfree(adap2pinfo(adapter, i)->rss);
4546 			free_netdev(adapter->port[i]);
4547 		}
4548 	if (adapter->flags & FW_OK)
4549 		t4_fw_bye(adapter, adapter->pf);
4550 }
4551 
4552 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4553 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4554 		   NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4555 #define SEGMENT_SIZE 128
4556 
4557 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4558 {
4559 	int func, i, err, s_qpp, qpp, num_seg;
4560 	struct port_info *pi;
4561 	bool highdma = false;
4562 	struct adapter *adapter = NULL;
4563 	void __iomem *regs;
4564 
4565 	printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4566 
4567 	err = pci_request_regions(pdev, KBUILD_MODNAME);
4568 	if (err) {
4569 		/* Just info, some other driver may have claimed the device. */
4570 		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4571 		return err;
4572 	}
4573 
4574 	err = pci_enable_device(pdev);
4575 	if (err) {
4576 		dev_err(&pdev->dev, "cannot enable PCI device\n");
4577 		goto out_release_regions;
4578 	}
4579 
4580 	regs = pci_ioremap_bar(pdev, 0);
4581 	if (!regs) {
4582 		dev_err(&pdev->dev, "cannot map device registers\n");
4583 		err = -ENOMEM;
4584 		goto out_disable_device;
4585 	}
4586 
4587 	err = t4_wait_dev_ready(regs);
4588 	if (err < 0)
4589 		goto out_unmap_bar0;
4590 
4591 	/* We control everything through one PF */
4592 	func = SOURCEPF_G(readl(regs + PL_WHOAMI_A));
4593 	if (func != ent->driver_data) {
4594 		iounmap(regs);
4595 		pci_disable_device(pdev);
4596 		pci_save_state(pdev);        /* to restore SR-IOV later */
4597 		goto sriov;
4598 	}
4599 
4600 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4601 		highdma = true;
4602 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4603 		if (err) {
4604 			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4605 				"coherent allocations\n");
4606 			goto out_unmap_bar0;
4607 		}
4608 	} else {
4609 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4610 		if (err) {
4611 			dev_err(&pdev->dev, "no usable DMA configuration\n");
4612 			goto out_unmap_bar0;
4613 		}
4614 	}
4615 
4616 	pci_enable_pcie_error_reporting(pdev);
4617 	enable_pcie_relaxed_ordering(pdev);
4618 	pci_set_master(pdev);
4619 	pci_save_state(pdev);
4620 
4621 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4622 	if (!adapter) {
4623 		err = -ENOMEM;
4624 		goto out_unmap_bar0;
4625 	}
4626 
4627 	adapter->workq = create_singlethread_workqueue("cxgb4");
4628 	if (!adapter->workq) {
4629 		err = -ENOMEM;
4630 		goto out_free_adapter;
4631 	}
4632 
4633 	/* PCI device has been enabled */
4634 	adapter->flags |= DEV_ENABLED;
4635 
4636 	adapter->regs = regs;
4637 	adapter->pdev = pdev;
4638 	adapter->pdev_dev = &pdev->dev;
4639 	adapter->mbox = func;
4640 	adapter->pf = func;
4641 	adapter->msg_enable = dflt_msg_enable;
4642 	memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4643 
4644 	spin_lock_init(&adapter->stats_lock);
4645 	spin_lock_init(&adapter->tid_release_lock);
4646 	spin_lock_init(&adapter->win0_lock);
4647 
4648 	INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
4649 	INIT_WORK(&adapter->db_full_task, process_db_full);
4650 	INIT_WORK(&adapter->db_drop_task, process_db_drop);
4651 
4652 	err = t4_prep_adapter(adapter);
4653 	if (err)
4654 		goto out_free_adapter;
4655 
4656 
4657 	if (!is_t4(adapter->params.chip)) {
4658 		s_qpp = (QUEUESPERPAGEPF0_S +
4659 			(QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
4660 			adapter->pf);
4661 		qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
4662 		      SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
4663 		num_seg = PAGE_SIZE / SEGMENT_SIZE;
4664 
4665 		/* Each segment size is 128B. Write coalescing is enabled only
4666 		 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
4667 		 * queue is less no of segments that can be accommodated in
4668 		 * a page size.
4669 		 */
4670 		if (qpp > num_seg) {
4671 			dev_err(&pdev->dev,
4672 				"Incorrect number of egress queues per page\n");
4673 			err = -EINVAL;
4674 			goto out_free_adapter;
4675 		}
4676 		adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
4677 		pci_resource_len(pdev, 2));
4678 		if (!adapter->bar2) {
4679 			dev_err(&pdev->dev, "cannot map device bar2 region\n");
4680 			err = -ENOMEM;
4681 			goto out_free_adapter;
4682 		}
4683 		t4_write_reg(adapter, SGE_STAT_CFG_A,
4684 			     STATSOURCE_T5_V(7) | STATMODE_V(0));
4685 	}
4686 
4687 	setup_memwin(adapter);
4688 	err = adap_init0(adapter);
4689 #ifdef CONFIG_DEBUG_FS
4690 	bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
4691 #endif
4692 	setup_memwin_rdma(adapter);
4693 	if (err)
4694 		goto out_unmap_bar;
4695 
4696 	for_each_port(adapter, i) {
4697 		struct net_device *netdev;
4698 
4699 		netdev = alloc_etherdev_mq(sizeof(struct port_info),
4700 					   MAX_ETH_QSETS);
4701 		if (!netdev) {
4702 			err = -ENOMEM;
4703 			goto out_free_dev;
4704 		}
4705 
4706 		SET_NETDEV_DEV(netdev, &pdev->dev);
4707 
4708 		adapter->port[i] = netdev;
4709 		pi = netdev_priv(netdev);
4710 		pi->adapter = adapter;
4711 		pi->xact_addr_filt = -1;
4712 		pi->port_id = i;
4713 		netdev->irq = pdev->irq;
4714 
4715 		netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
4716 			NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4717 			NETIF_F_RXCSUM | NETIF_F_RXHASH |
4718 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
4719 		if (highdma)
4720 			netdev->hw_features |= NETIF_F_HIGHDMA;
4721 		netdev->features |= netdev->hw_features;
4722 		netdev->vlan_features = netdev->features & VLAN_FEAT;
4723 
4724 		netdev->priv_flags |= IFF_UNICAST_FLT;
4725 
4726 		netdev->netdev_ops = &cxgb4_netdev_ops;
4727 #ifdef CONFIG_CHELSIO_T4_DCB
4728 		netdev->dcbnl_ops = &cxgb4_dcb_ops;
4729 		cxgb4_dcb_state_init(netdev);
4730 #endif
4731 		cxgb4_set_ethtool_ops(netdev);
4732 	}
4733 
4734 	pci_set_drvdata(pdev, adapter);
4735 
4736 	if (adapter->flags & FW_OK) {
4737 		err = t4_port_init(adapter, func, func, 0);
4738 		if (err)
4739 			goto out_free_dev;
4740 	} else if (adapter->params.nports == 1) {
4741 		/* If we don't have a connection to the firmware -- possibly
4742 		 * because of an error -- grab the raw VPD parameters so we
4743 		 * can set the proper MAC Address on the debug network
4744 		 * interface that we've created.
4745 		 */
4746 		u8 hw_addr[ETH_ALEN];
4747 		u8 *na = adapter->params.vpd.na;
4748 
4749 		err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
4750 		if (!err) {
4751 			for (i = 0; i < ETH_ALEN; i++)
4752 				hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
4753 					      hex2val(na[2 * i + 1]));
4754 			t4_set_hw_addr(adapter, 0, hw_addr);
4755 		}
4756 	}
4757 
4758 	/* Configure queues and allocate tables now, they can be needed as
4759 	 * soon as the first register_netdev completes.
4760 	 */
4761 	cfg_queues(adapter);
4762 
4763 	adapter->l2t = t4_init_l2t();
4764 	if (!adapter->l2t) {
4765 		/* We tolerate a lack of L2T, giving up some functionality */
4766 		dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
4767 		adapter->params.offload = 0;
4768 	}
4769 
4770 #if IS_ENABLED(CONFIG_IPV6)
4771 	adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
4772 					  adapter->clipt_end);
4773 	if (!adapter->clipt) {
4774 		/* We tolerate a lack of clip_table, giving up
4775 		 * some functionality
4776 		 */
4777 		dev_warn(&pdev->dev,
4778 			 "could not allocate Clip table, continuing\n");
4779 		adapter->params.offload = 0;
4780 	}
4781 #endif
4782 	if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
4783 		dev_warn(&pdev->dev, "could not allocate TID table, "
4784 			 "continuing\n");
4785 		adapter->params.offload = 0;
4786 	}
4787 
4788 	/* See what interrupts we'll be using */
4789 	if (msi > 1 && enable_msix(adapter) == 0)
4790 		adapter->flags |= USING_MSIX;
4791 	else if (msi > 0 && pci_enable_msi(pdev) == 0)
4792 		adapter->flags |= USING_MSI;
4793 
4794 	err = init_rss(adapter);
4795 	if (err)
4796 		goto out_free_dev;
4797 
4798 	/*
4799 	 * The card is now ready to go.  If any errors occur during device
4800 	 * registration we do not fail the whole card but rather proceed only
4801 	 * with the ports we manage to register successfully.  However we must
4802 	 * register at least one net device.
4803 	 */
4804 	for_each_port(adapter, i) {
4805 		pi = adap2pinfo(adapter, i);
4806 		netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
4807 		netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
4808 
4809 		err = register_netdev(adapter->port[i]);
4810 		if (err)
4811 			break;
4812 		adapter->chan_map[pi->tx_chan] = i;
4813 		print_port_info(adapter->port[i]);
4814 	}
4815 	if (i == 0) {
4816 		dev_err(&pdev->dev, "could not register any net devices\n");
4817 		goto out_free_dev;
4818 	}
4819 	if (err) {
4820 		dev_warn(&pdev->dev, "only %d net devices registered\n", i);
4821 		err = 0;
4822 	}
4823 
4824 	if (cxgb4_debugfs_root) {
4825 		adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
4826 							   cxgb4_debugfs_root);
4827 		setup_debugfs(adapter);
4828 	}
4829 
4830 	/* PCIe EEH recovery on powerpc platforms needs fundamental reset */
4831 	pdev->needs_freset = 1;
4832 
4833 	if (is_offload(adapter))
4834 		attach_ulds(adapter);
4835 
4836 sriov:
4837 #ifdef CONFIG_PCI_IOV
4838 	if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
4839 		if (pci_enable_sriov(pdev, num_vf[func]) == 0)
4840 			dev_info(&pdev->dev,
4841 				 "instantiated %u virtual functions\n",
4842 				 num_vf[func]);
4843 #endif
4844 	return 0;
4845 
4846  out_free_dev:
4847 	free_some_resources(adapter);
4848  out_unmap_bar:
4849 	if (!is_t4(adapter->params.chip))
4850 		iounmap(adapter->bar2);
4851  out_free_adapter:
4852 	if (adapter->workq)
4853 		destroy_workqueue(adapter->workq);
4854 
4855 	kfree(adapter);
4856  out_unmap_bar0:
4857 	iounmap(regs);
4858  out_disable_device:
4859 	pci_disable_pcie_error_reporting(pdev);
4860 	pci_disable_device(pdev);
4861  out_release_regions:
4862 	pci_release_regions(pdev);
4863 	return err;
4864 }
4865 
4866 static void remove_one(struct pci_dev *pdev)
4867 {
4868 	struct adapter *adapter = pci_get_drvdata(pdev);
4869 
4870 #ifdef CONFIG_PCI_IOV
4871 	pci_disable_sriov(pdev);
4872 
4873 #endif
4874 
4875 	if (adapter) {
4876 		int i;
4877 
4878 		/* Tear down per-adapter Work Queue first since it can contain
4879 		 * references to our adapter data structure.
4880 		 */
4881 		destroy_workqueue(adapter->workq);
4882 
4883 		if (is_offload(adapter))
4884 			detach_ulds(adapter);
4885 
4886 		disable_interrupts(adapter);
4887 
4888 		for_each_port(adapter, i)
4889 			if (adapter->port[i]->reg_state == NETREG_REGISTERED)
4890 				unregister_netdev(adapter->port[i]);
4891 
4892 		debugfs_remove_recursive(adapter->debugfs_root);
4893 
4894 		/* If we allocated filters, free up state associated with any
4895 		 * valid filters ...
4896 		 */
4897 		if (adapter->tids.ftid_tab) {
4898 			struct filter_entry *f = &adapter->tids.ftid_tab[0];
4899 			for (i = 0; i < (adapter->tids.nftids +
4900 					adapter->tids.nsftids); i++, f++)
4901 				if (f->valid)
4902 					clear_filter(adapter, f);
4903 		}
4904 
4905 		if (adapter->flags & FULL_INIT_DONE)
4906 			cxgb_down(adapter);
4907 
4908 		free_some_resources(adapter);
4909 #if IS_ENABLED(CONFIG_IPV6)
4910 		t4_cleanup_clip_tbl(adapter);
4911 #endif
4912 		iounmap(adapter->regs);
4913 		if (!is_t4(adapter->params.chip))
4914 			iounmap(adapter->bar2);
4915 		pci_disable_pcie_error_reporting(pdev);
4916 		if ((adapter->flags & DEV_ENABLED)) {
4917 			pci_disable_device(pdev);
4918 			adapter->flags &= ~DEV_ENABLED;
4919 		}
4920 		pci_release_regions(pdev);
4921 		synchronize_rcu();
4922 		kfree(adapter);
4923 	} else
4924 		pci_release_regions(pdev);
4925 }
4926 
4927 static struct pci_driver cxgb4_driver = {
4928 	.name     = KBUILD_MODNAME,
4929 	.id_table = cxgb4_pci_tbl,
4930 	.probe    = init_one,
4931 	.remove   = remove_one,
4932 	.shutdown = remove_one,
4933 	.err_handler = &cxgb4_eeh,
4934 };
4935 
4936 static int __init cxgb4_init_module(void)
4937 {
4938 	int ret;
4939 
4940 	/* Debugfs support is optional, just warn if this fails */
4941 	cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
4942 	if (!cxgb4_debugfs_root)
4943 		pr_warn("could not create debugfs entry, continuing\n");
4944 
4945 	ret = pci_register_driver(&cxgb4_driver);
4946 	if (ret < 0)
4947 		debugfs_remove(cxgb4_debugfs_root);
4948 
4949 #if IS_ENABLED(CONFIG_IPV6)
4950 	if (!inet6addr_registered) {
4951 		register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
4952 		inet6addr_registered = true;
4953 	}
4954 #endif
4955 
4956 	return ret;
4957 }
4958 
4959 static void __exit cxgb4_cleanup_module(void)
4960 {
4961 #if IS_ENABLED(CONFIG_IPV6)
4962 	if (inet6addr_registered) {
4963 		unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
4964 		inet6addr_registered = false;
4965 	}
4966 #endif
4967 	pci_unregister_driver(&cxgb4_driver);
4968 	debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
4969 }
4970 
4971 module_init(cxgb4_init_module);
4972 module_exit(cxgb4_cleanup_module);
4973