1 /*
2  * This file is part of the Chelsio T4 Ethernet driver for Linux.
3  *
4  * Copyright (c) 2003-2010 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 <asm/uaccess.h>
65 
66 #include "cxgb4.h"
67 #include "t4_regs.h"
68 #include "t4_msg.h"
69 #include "t4fw_api.h"
70 #include "l2t.h"
71 
72 #include <../drivers/net/bonding/bonding.h>
73 
74 #ifdef DRV_VERSION
75 #undef DRV_VERSION
76 #endif
77 #define DRV_VERSION "2.0.0-ko"
78 #define DRV_DESC "Chelsio T4/T5 Network Driver"
79 
80 /*
81  * Max interrupt hold-off timer value in us.  Queues fall back to this value
82  * under extreme memory pressure so it's largish to give the system time to
83  * recover.
84  */
85 #define MAX_SGE_TIMERVAL 200U
86 
87 enum {
88 	/*
89 	 * Physical Function provisioning constants.
90 	 */
91 	PFRES_NVI = 4,			/* # of Virtual Interfaces */
92 	PFRES_NETHCTRL = 128,		/* # of EQs used for ETH or CTRL Qs */
93 	PFRES_NIQFLINT = 128,		/* # of ingress Qs/w Free List(s)/intr
94 					 */
95 	PFRES_NEQ = 256,		/* # of egress queues */
96 	PFRES_NIQ = 0,			/* # of ingress queues */
97 	PFRES_TC = 0,			/* PCI-E traffic class */
98 	PFRES_NEXACTF = 128,		/* # of exact MPS filters */
99 
100 	PFRES_R_CAPS = FW_CMD_CAP_PF,
101 	PFRES_WX_CAPS = FW_CMD_CAP_PF,
102 
103 #ifdef CONFIG_PCI_IOV
104 	/*
105 	 * Virtual Function provisioning constants.  We need two extra Ingress
106 	 * Queues with Interrupt capability to serve as the VF's Firmware
107 	 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
108 	 * neither will have Free Lists associated with them).  For each
109 	 * Ethernet/Control Egress Queue and for each Free List, we need an
110 	 * Egress Context.
111 	 */
112 	VFRES_NPORTS = 1,		/* # of "ports" per VF */
113 	VFRES_NQSETS = 2,		/* # of "Queue Sets" per VF */
114 
115 	VFRES_NVI = VFRES_NPORTS,	/* # of Virtual Interfaces */
116 	VFRES_NETHCTRL = VFRES_NQSETS,	/* # of EQs used for ETH or CTRL Qs */
117 	VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
118 	VFRES_NEQ = VFRES_NQSETS*2,	/* # of egress queues */
119 	VFRES_NIQ = 0,			/* # of non-fl/int ingress queues */
120 	VFRES_TC = 0,			/* PCI-E traffic class */
121 	VFRES_NEXACTF = 16,		/* # of exact MPS filters */
122 
123 	VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
124 	VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
125 #endif
126 };
127 
128 /*
129  * Provide a Port Access Rights Mask for the specified PF/VF.  This is very
130  * static and likely not to be useful in the long run.  We really need to
131  * implement some form of persistent configuration which the firmware
132  * controls.
133  */
134 static unsigned int pfvfres_pmask(struct adapter *adapter,
135 				  unsigned int pf, unsigned int vf)
136 {
137 	unsigned int portn, portvec;
138 
139 	/*
140 	 * Give PF's access to all of the ports.
141 	 */
142 	if (vf == 0)
143 		return FW_PFVF_CMD_PMASK_MASK;
144 
145 	/*
146 	 * For VFs, we'll assign them access to the ports based purely on the
147 	 * PF.  We assign active ports in order, wrapping around if there are
148 	 * fewer active ports than PFs: e.g. active port[pf % nports].
149 	 * Unfortunately the adapter's port_info structs haven't been
150 	 * initialized yet so we have to compute this.
151 	 */
152 	if (adapter->params.nports == 0)
153 		return 0;
154 
155 	portn = pf % adapter->params.nports;
156 	portvec = adapter->params.portvec;
157 	for (;;) {
158 		/*
159 		 * Isolate the lowest set bit in the port vector.  If we're at
160 		 * the port number that we want, return that as the pmask.
161 		 * otherwise mask that bit out of the port vector and
162 		 * decrement our port number ...
163 		 */
164 		unsigned int pmask = portvec ^ (portvec & (portvec-1));
165 		if (portn == 0)
166 			return pmask;
167 		portn--;
168 		portvec &= ~pmask;
169 	}
170 	/*NOTREACHED*/
171 }
172 
173 enum {
174 	MAX_TXQ_ENTRIES      = 16384,
175 	MAX_CTRL_TXQ_ENTRIES = 1024,
176 	MAX_RSPQ_ENTRIES     = 16384,
177 	MAX_RX_BUFFERS       = 16384,
178 	MIN_TXQ_ENTRIES      = 32,
179 	MIN_CTRL_TXQ_ENTRIES = 32,
180 	MIN_RSPQ_ENTRIES     = 128,
181 	MIN_FL_ENTRIES       = 16
182 };
183 
184 /* Host shadow copy of ingress filter entry.  This is in host native format
185  * and doesn't match the ordering or bit order, etc. of the hardware of the
186  * firmware command.  The use of bit-field structure elements is purely to
187  * remind ourselves of the field size limitations and save memory in the case
188  * where the filter table is large.
189  */
190 struct filter_entry {
191 	/* Administrative fields for filter.
192 	 */
193 	u32 valid:1;            /* filter allocated and valid */
194 	u32 locked:1;           /* filter is administratively locked */
195 
196 	u32 pending:1;          /* filter action is pending firmware reply */
197 	u32 smtidx:8;           /* Source MAC Table index for smac */
198 	struct l2t_entry *l2t;  /* Layer Two Table entry for dmac */
199 
200 	/* The filter itself.  Most of this is a straight copy of information
201 	 * provided by the extended ioctl().  Some fields are translated to
202 	 * internal forms -- for instance the Ingress Queue ID passed in from
203 	 * the ioctl() is translated into the Absolute Ingress Queue ID.
204 	 */
205 	struct ch_filter_specification fs;
206 };
207 
208 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
209 			 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
210 			 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
211 
212 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
213 
214 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
215 	CH_DEVICE(0xa000, 0),  /* PE10K */
216 	CH_DEVICE(0x4001, -1),
217 	CH_DEVICE(0x4002, -1),
218 	CH_DEVICE(0x4003, -1),
219 	CH_DEVICE(0x4004, -1),
220 	CH_DEVICE(0x4005, -1),
221 	CH_DEVICE(0x4006, -1),
222 	CH_DEVICE(0x4007, -1),
223 	CH_DEVICE(0x4008, -1),
224 	CH_DEVICE(0x4009, -1),
225 	CH_DEVICE(0x400a, -1),
226 	CH_DEVICE(0x4401, 4),
227 	CH_DEVICE(0x4402, 4),
228 	CH_DEVICE(0x4403, 4),
229 	CH_DEVICE(0x4404, 4),
230 	CH_DEVICE(0x4405, 4),
231 	CH_DEVICE(0x4406, 4),
232 	CH_DEVICE(0x4407, 4),
233 	CH_DEVICE(0x4408, 4),
234 	CH_DEVICE(0x4409, 4),
235 	CH_DEVICE(0x440a, 4),
236 	CH_DEVICE(0x440d, 4),
237 	CH_DEVICE(0x440e, 4),
238 	CH_DEVICE(0x5001, 4),
239 	CH_DEVICE(0x5002, 4),
240 	CH_DEVICE(0x5003, 4),
241 	CH_DEVICE(0x5004, 4),
242 	CH_DEVICE(0x5005, 4),
243 	CH_DEVICE(0x5006, 4),
244 	CH_DEVICE(0x5007, 4),
245 	CH_DEVICE(0x5008, 4),
246 	CH_DEVICE(0x5009, 4),
247 	CH_DEVICE(0x500A, 4),
248 	CH_DEVICE(0x500B, 4),
249 	CH_DEVICE(0x500C, 4),
250 	CH_DEVICE(0x500D, 4),
251 	CH_DEVICE(0x500E, 4),
252 	CH_DEVICE(0x500F, 4),
253 	CH_DEVICE(0x5010, 4),
254 	CH_DEVICE(0x5011, 4),
255 	CH_DEVICE(0x5012, 4),
256 	CH_DEVICE(0x5013, 4),
257 	CH_DEVICE(0x5401, 4),
258 	CH_DEVICE(0x5402, 4),
259 	CH_DEVICE(0x5403, 4),
260 	CH_DEVICE(0x5404, 4),
261 	CH_DEVICE(0x5405, 4),
262 	CH_DEVICE(0x5406, 4),
263 	CH_DEVICE(0x5407, 4),
264 	CH_DEVICE(0x5408, 4),
265 	CH_DEVICE(0x5409, 4),
266 	CH_DEVICE(0x540A, 4),
267 	CH_DEVICE(0x540B, 4),
268 	CH_DEVICE(0x540C, 4),
269 	CH_DEVICE(0x540D, 4),
270 	CH_DEVICE(0x540E, 4),
271 	CH_DEVICE(0x540F, 4),
272 	CH_DEVICE(0x5410, 4),
273 	CH_DEVICE(0x5411, 4),
274 	CH_DEVICE(0x5412, 4),
275 	CH_DEVICE(0x5413, 4),
276 	{ 0, }
277 };
278 
279 #define FW_FNAME "cxgb4/t4fw.bin"
280 #define FW5_FNAME "cxgb4/t5fw.bin"
281 #define FW_CFNAME "cxgb4/t4-config.txt"
282 #define FW5_CFNAME "cxgb4/t5-config.txt"
283 
284 MODULE_DESCRIPTION(DRV_DESC);
285 MODULE_AUTHOR("Chelsio Communications");
286 MODULE_LICENSE("Dual BSD/GPL");
287 MODULE_VERSION(DRV_VERSION);
288 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
289 MODULE_FIRMWARE(FW_FNAME);
290 MODULE_FIRMWARE(FW5_FNAME);
291 
292 /*
293  * Normally we're willing to become the firmware's Master PF but will be happy
294  * if another PF has already become the Master and initialized the adapter.
295  * Setting "force_init" will cause this driver to forcibly establish itself as
296  * the Master PF and initialize the adapter.
297  */
298 static uint force_init;
299 
300 module_param(force_init, uint, 0644);
301 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
302 
303 /*
304  * Normally if the firmware we connect to has Configuration File support, we
305  * use that and only fall back to the old Driver-based initialization if the
306  * Configuration File fails for some reason.  If force_old_init is set, then
307  * we'll always use the old Driver-based initialization sequence.
308  */
309 static uint force_old_init;
310 
311 module_param(force_old_init, uint, 0644);
312 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
313 
314 static int dflt_msg_enable = DFLT_MSG_ENABLE;
315 
316 module_param(dflt_msg_enable, int, 0644);
317 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
318 
319 /*
320  * The driver uses the best interrupt scheme available on a platform in the
321  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
322  * of these schemes the driver may consider as follows:
323  *
324  * msi = 2: choose from among all three options
325  * msi = 1: only consider MSI and INTx interrupts
326  * msi = 0: force INTx interrupts
327  */
328 static int msi = 2;
329 
330 module_param(msi, int, 0644);
331 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
332 
333 /*
334  * Queue interrupt hold-off timer values.  Queues default to the first of these
335  * upon creation.
336  */
337 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
338 
339 module_param_array(intr_holdoff, uint, NULL, 0644);
340 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
341 		 "0..4 in microseconds");
342 
343 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
344 
345 module_param_array(intr_cnt, uint, NULL, 0644);
346 MODULE_PARM_DESC(intr_cnt,
347 		 "thresholds 1..3 for queue interrupt packet counters");
348 
349 /*
350  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
351  * offset by 2 bytes in order to have the IP headers line up on 4-byte
352  * boundaries.  This is a requirement for many architectures which will throw
353  * a machine check fault if an attempt is made to access one of the 4-byte IP
354  * header fields on a non-4-byte boundary.  And it's a major performance issue
355  * even on some architectures which allow it like some implementations of the
356  * x86 ISA.  However, some architectures don't mind this and for some very
357  * edge-case performance sensitive applications (like forwarding large volumes
358  * of small packets), setting this DMA offset to 0 will decrease the number of
359  * PCI-E Bus transfers enough to measurably affect performance.
360  */
361 static int rx_dma_offset = 2;
362 
363 static bool vf_acls;
364 
365 #ifdef CONFIG_PCI_IOV
366 module_param(vf_acls, bool, 0644);
367 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
368 
369 /* Configure the number of PCI-E Virtual Function which are to be instantiated
370  * on SR-IOV Capable Physical Functions.
371  */
372 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV];
373 
374 module_param_array(num_vf, uint, NULL, 0644);
375 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
376 #endif
377 
378 /*
379  * The filter TCAM has a fixed portion and a variable portion.  The fixed
380  * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
381  * ports.  The variable portion is 36 bits which can include things like Exact
382  * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
383  * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
384  * far exceed the 36-bit budget for this "compressed" header portion of the
385  * filter.  Thus, we have a scarce resource which must be carefully managed.
386  *
387  * By default we set this up to mostly match the set of filter matching
388  * capabilities of T3 but with accommodations for some of T4's more
389  * interesting features:
390  *
391  *   { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
392  *     [Inner] VLAN (17), Port (3), FCoE (1) }
393  */
394 enum {
395 	TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
396 	TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
397 	TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
398 };
399 
400 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
401 
402 module_param(tp_vlan_pri_map, uint, 0644);
403 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
404 
405 static struct dentry *cxgb4_debugfs_root;
406 
407 static LIST_HEAD(adapter_list);
408 static DEFINE_MUTEX(uld_mutex);
409 /* Adapter list to be accessed from atomic context */
410 static LIST_HEAD(adap_rcu_list);
411 static DEFINE_SPINLOCK(adap_rcu_lock);
412 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
413 static const char *uld_str[] = { "RDMA", "iSCSI" };
414 
415 static void link_report(struct net_device *dev)
416 {
417 	if (!netif_carrier_ok(dev))
418 		netdev_info(dev, "link down\n");
419 	else {
420 		static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
421 
422 		const char *s = "10Mbps";
423 		const struct port_info *p = netdev_priv(dev);
424 
425 		switch (p->link_cfg.speed) {
426 		case SPEED_10000:
427 			s = "10Gbps";
428 			break;
429 		case SPEED_1000:
430 			s = "1000Mbps";
431 			break;
432 		case SPEED_100:
433 			s = "100Mbps";
434 			break;
435 		}
436 
437 		netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
438 			    fc[p->link_cfg.fc]);
439 	}
440 }
441 
442 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
443 {
444 	struct net_device *dev = adapter->port[port_id];
445 
446 	/* Skip changes from disabled ports. */
447 	if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
448 		if (link_stat)
449 			netif_carrier_on(dev);
450 		else
451 			netif_carrier_off(dev);
452 
453 		link_report(dev);
454 	}
455 }
456 
457 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
458 {
459 	static const char *mod_str[] = {
460 		NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
461 	};
462 
463 	const struct net_device *dev = adap->port[port_id];
464 	const struct port_info *pi = netdev_priv(dev);
465 
466 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
467 		netdev_info(dev, "port module unplugged\n");
468 	else if (pi->mod_type < ARRAY_SIZE(mod_str))
469 		netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
470 }
471 
472 /*
473  * Configure the exact and hash address filters to handle a port's multicast
474  * and secondary unicast MAC addresses.
475  */
476 static int set_addr_filters(const struct net_device *dev, bool sleep)
477 {
478 	u64 mhash = 0;
479 	u64 uhash = 0;
480 	bool free = true;
481 	u16 filt_idx[7];
482 	const u8 *addr[7];
483 	int ret, naddr = 0;
484 	const struct netdev_hw_addr *ha;
485 	int uc_cnt = netdev_uc_count(dev);
486 	int mc_cnt = netdev_mc_count(dev);
487 	const struct port_info *pi = netdev_priv(dev);
488 	unsigned int mb = pi->adapter->fn;
489 
490 	/* first do the secondary unicast addresses */
491 	netdev_for_each_uc_addr(ha, dev) {
492 		addr[naddr++] = ha->addr;
493 		if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
494 			ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
495 					naddr, addr, filt_idx, &uhash, sleep);
496 			if (ret < 0)
497 				return ret;
498 
499 			free = false;
500 			naddr = 0;
501 		}
502 	}
503 
504 	/* next set up the multicast addresses */
505 	netdev_for_each_mc_addr(ha, dev) {
506 		addr[naddr++] = ha->addr;
507 		if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
508 			ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
509 					naddr, addr, filt_idx, &mhash, sleep);
510 			if (ret < 0)
511 				return ret;
512 
513 			free = false;
514 			naddr = 0;
515 		}
516 	}
517 
518 	return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
519 				uhash | mhash, sleep);
520 }
521 
522 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
523 module_param(dbfifo_int_thresh, int, 0644);
524 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
525 
526 /*
527  * usecs to sleep while draining the dbfifo
528  */
529 static int dbfifo_drain_delay = 1000;
530 module_param(dbfifo_drain_delay, int, 0644);
531 MODULE_PARM_DESC(dbfifo_drain_delay,
532 		 "usecs to sleep while draining the dbfifo");
533 
534 /*
535  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
536  * If @mtu is -1 it is left unchanged.
537  */
538 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
539 {
540 	int ret;
541 	struct port_info *pi = netdev_priv(dev);
542 
543 	ret = set_addr_filters(dev, sleep_ok);
544 	if (ret == 0)
545 		ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
546 				    (dev->flags & IFF_PROMISC) ? 1 : 0,
547 				    (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
548 				    sleep_ok);
549 	return ret;
550 }
551 
552 static struct workqueue_struct *workq;
553 
554 /**
555  *	link_start - enable a port
556  *	@dev: the port to enable
557  *
558  *	Performs the MAC and PHY actions needed to enable a port.
559  */
560 static int link_start(struct net_device *dev)
561 {
562 	int ret;
563 	struct port_info *pi = netdev_priv(dev);
564 	unsigned int mb = pi->adapter->fn;
565 
566 	/*
567 	 * We do not set address filters and promiscuity here, the stack does
568 	 * that step explicitly.
569 	 */
570 	ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
571 			    !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
572 	if (ret == 0) {
573 		ret = t4_change_mac(pi->adapter, mb, pi->viid,
574 				    pi->xact_addr_filt, dev->dev_addr, true,
575 				    true);
576 		if (ret >= 0) {
577 			pi->xact_addr_filt = ret;
578 			ret = 0;
579 		}
580 	}
581 	if (ret == 0)
582 		ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
583 				    &pi->link_cfg);
584 	if (ret == 0)
585 		ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
586 	return ret;
587 }
588 
589 /* Clear a filter and release any of its resources that we own.  This also
590  * clears the filter's "pending" status.
591  */
592 static void clear_filter(struct adapter *adap, struct filter_entry *f)
593 {
594 	/* If the new or old filter have loopback rewriteing rules then we'll
595 	 * need to free any existing Layer Two Table (L2T) entries of the old
596 	 * filter rule.  The firmware will handle freeing up any Source MAC
597 	 * Table (SMT) entries used for rewriting Source MAC Addresses in
598 	 * loopback rules.
599 	 */
600 	if (f->l2t)
601 		cxgb4_l2t_release(f->l2t);
602 
603 	/* The zeroing of the filter rule below clears the filter valid,
604 	 * pending, locked flags, l2t pointer, etc. so it's all we need for
605 	 * this operation.
606 	 */
607 	memset(f, 0, sizeof(*f));
608 }
609 
610 /* Handle a filter write/deletion reply.
611  */
612 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
613 {
614 	unsigned int idx = GET_TID(rpl);
615 	unsigned int nidx = idx - adap->tids.ftid_base;
616 	unsigned int ret;
617 	struct filter_entry *f;
618 
619 	if (idx >= adap->tids.ftid_base && nidx <
620 	   (adap->tids.nftids + adap->tids.nsftids)) {
621 		idx = nidx;
622 		ret = GET_TCB_COOKIE(rpl->cookie);
623 		f = &adap->tids.ftid_tab[idx];
624 
625 		if (ret == FW_FILTER_WR_FLT_DELETED) {
626 			/* Clear the filter when we get confirmation from the
627 			 * hardware that the filter has been deleted.
628 			 */
629 			clear_filter(adap, f);
630 		} else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
631 			dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
632 				idx);
633 			clear_filter(adap, f);
634 		} else if (ret == FW_FILTER_WR_FLT_ADDED) {
635 			f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
636 			f->pending = 0;  /* asynchronous setup completed */
637 			f->valid = 1;
638 		} else {
639 			/* Something went wrong.  Issue a warning about the
640 			 * problem and clear everything out.
641 			 */
642 			dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
643 				idx, ret);
644 			clear_filter(adap, f);
645 		}
646 	}
647 }
648 
649 /* Response queue handler for the FW event queue.
650  */
651 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
652 			  const struct pkt_gl *gl)
653 {
654 	u8 opcode = ((const struct rss_header *)rsp)->opcode;
655 
656 	rsp++;                                          /* skip RSS header */
657 
658 	/* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
659 	 */
660 	if (unlikely(opcode == CPL_FW4_MSG &&
661 	   ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
662 		rsp++;
663 		opcode = ((const struct rss_header *)rsp)->opcode;
664 		rsp++;
665 		if (opcode != CPL_SGE_EGR_UPDATE) {
666 			dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
667 				, opcode);
668 			goto out;
669 		}
670 	}
671 
672 	if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
673 		const struct cpl_sge_egr_update *p = (void *)rsp;
674 		unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
675 		struct sge_txq *txq;
676 
677 		txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
678 		txq->restarts++;
679 		if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
680 			struct sge_eth_txq *eq;
681 
682 			eq = container_of(txq, struct sge_eth_txq, q);
683 			netif_tx_wake_queue(eq->txq);
684 		} else {
685 			struct sge_ofld_txq *oq;
686 
687 			oq = container_of(txq, struct sge_ofld_txq, q);
688 			tasklet_schedule(&oq->qresume_tsk);
689 		}
690 	} else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
691 		const struct cpl_fw6_msg *p = (void *)rsp;
692 
693 		if (p->type == 0)
694 			t4_handle_fw_rpl(q->adap, p->data);
695 	} else if (opcode == CPL_L2T_WRITE_RPL) {
696 		const struct cpl_l2t_write_rpl *p = (void *)rsp;
697 
698 		do_l2t_write_rpl(q->adap, p);
699 	} else if (opcode == CPL_SET_TCB_RPL) {
700 		const struct cpl_set_tcb_rpl *p = (void *)rsp;
701 
702 		filter_rpl(q->adap, p);
703 	} else
704 		dev_err(q->adap->pdev_dev,
705 			"unexpected CPL %#x on FW event queue\n", opcode);
706 out:
707 	return 0;
708 }
709 
710 /**
711  *	uldrx_handler - response queue handler for ULD queues
712  *	@q: the response queue that received the packet
713  *	@rsp: the response queue descriptor holding the offload message
714  *	@gl: the gather list of packet fragments
715  *
716  *	Deliver an ingress offload packet to a ULD.  All processing is done by
717  *	the ULD, we just maintain statistics.
718  */
719 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
720 			 const struct pkt_gl *gl)
721 {
722 	struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
723 
724 	/* FW can send CPLs encapsulated in a CPL_FW4_MSG.
725 	 */
726 	if (((const struct rss_header *)rsp)->opcode == CPL_FW4_MSG &&
727 	    ((const struct cpl_fw4_msg *)(rsp + 1))->type == FW_TYPE_RSSCPL)
728 		rsp += 2;
729 
730 	if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
731 		rxq->stats.nomem++;
732 		return -1;
733 	}
734 	if (gl == NULL)
735 		rxq->stats.imm++;
736 	else if (gl == CXGB4_MSG_AN)
737 		rxq->stats.an++;
738 	else
739 		rxq->stats.pkts++;
740 	return 0;
741 }
742 
743 static void disable_msi(struct adapter *adapter)
744 {
745 	if (adapter->flags & USING_MSIX) {
746 		pci_disable_msix(adapter->pdev);
747 		adapter->flags &= ~USING_MSIX;
748 	} else if (adapter->flags & USING_MSI) {
749 		pci_disable_msi(adapter->pdev);
750 		adapter->flags &= ~USING_MSI;
751 	}
752 }
753 
754 /*
755  * Interrupt handler for non-data events used with MSI-X.
756  */
757 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
758 {
759 	struct adapter *adap = cookie;
760 
761 	u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
762 	if (v & PFSW) {
763 		adap->swintr = 1;
764 		t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
765 	}
766 	t4_slow_intr_handler(adap);
767 	return IRQ_HANDLED;
768 }
769 
770 /*
771  * Name the MSI-X interrupts.
772  */
773 static void name_msix_vecs(struct adapter *adap)
774 {
775 	int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
776 
777 	/* non-data interrupts */
778 	snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
779 
780 	/* FW events */
781 	snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
782 		 adap->port[0]->name);
783 
784 	/* Ethernet queues */
785 	for_each_port(adap, j) {
786 		struct net_device *d = adap->port[j];
787 		const struct port_info *pi = netdev_priv(d);
788 
789 		for (i = 0; i < pi->nqsets; i++, msi_idx++)
790 			snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
791 				 d->name, i);
792 	}
793 
794 	/* offload queues */
795 	for_each_ofldrxq(&adap->sge, i)
796 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
797 			 adap->port[0]->name, i);
798 
799 	for_each_rdmarxq(&adap->sge, i)
800 		snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
801 			 adap->port[0]->name, i);
802 }
803 
804 static int request_msix_queue_irqs(struct adapter *adap)
805 {
806 	struct sge *s = &adap->sge;
807 	int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
808 
809 	err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
810 			  adap->msix_info[1].desc, &s->fw_evtq);
811 	if (err)
812 		return err;
813 
814 	for_each_ethrxq(s, ethqidx) {
815 		err = request_irq(adap->msix_info[msi_index].vec,
816 				  t4_sge_intr_msix, 0,
817 				  adap->msix_info[msi_index].desc,
818 				  &s->ethrxq[ethqidx].rspq);
819 		if (err)
820 			goto unwind;
821 		msi_index++;
822 	}
823 	for_each_ofldrxq(s, ofldqidx) {
824 		err = request_irq(adap->msix_info[msi_index].vec,
825 				  t4_sge_intr_msix, 0,
826 				  adap->msix_info[msi_index].desc,
827 				  &s->ofldrxq[ofldqidx].rspq);
828 		if (err)
829 			goto unwind;
830 		msi_index++;
831 	}
832 	for_each_rdmarxq(s, rdmaqidx) {
833 		err = request_irq(adap->msix_info[msi_index].vec,
834 				  t4_sge_intr_msix, 0,
835 				  adap->msix_info[msi_index].desc,
836 				  &s->rdmarxq[rdmaqidx].rspq);
837 		if (err)
838 			goto unwind;
839 		msi_index++;
840 	}
841 	return 0;
842 
843 unwind:
844 	while (--rdmaqidx >= 0)
845 		free_irq(adap->msix_info[--msi_index].vec,
846 			 &s->rdmarxq[rdmaqidx].rspq);
847 	while (--ofldqidx >= 0)
848 		free_irq(adap->msix_info[--msi_index].vec,
849 			 &s->ofldrxq[ofldqidx].rspq);
850 	while (--ethqidx >= 0)
851 		free_irq(adap->msix_info[--msi_index].vec,
852 			 &s->ethrxq[ethqidx].rspq);
853 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
854 	return err;
855 }
856 
857 static void free_msix_queue_irqs(struct adapter *adap)
858 {
859 	int i, msi_index = 2;
860 	struct sge *s = &adap->sge;
861 
862 	free_irq(adap->msix_info[1].vec, &s->fw_evtq);
863 	for_each_ethrxq(s, i)
864 		free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
865 	for_each_ofldrxq(s, i)
866 		free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
867 	for_each_rdmarxq(s, i)
868 		free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
869 }
870 
871 /**
872  *	write_rss - write the RSS table for a given port
873  *	@pi: the port
874  *	@queues: array of queue indices for RSS
875  *
876  *	Sets up the portion of the HW RSS table for the port's VI to distribute
877  *	packets to the Rx queues in @queues.
878  */
879 static int write_rss(const struct port_info *pi, const u16 *queues)
880 {
881 	u16 *rss;
882 	int i, err;
883 	const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
884 
885 	rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
886 	if (!rss)
887 		return -ENOMEM;
888 
889 	/* map the queue indices to queue ids */
890 	for (i = 0; i < pi->rss_size; i++, queues++)
891 		rss[i] = q[*queues].rspq.abs_id;
892 
893 	err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
894 				  pi->rss_size, rss, pi->rss_size);
895 	kfree(rss);
896 	return err;
897 }
898 
899 /**
900  *	setup_rss - configure RSS
901  *	@adap: the adapter
902  *
903  *	Sets up RSS for each port.
904  */
905 static int setup_rss(struct adapter *adap)
906 {
907 	int i, err;
908 
909 	for_each_port(adap, i) {
910 		const struct port_info *pi = adap2pinfo(adap, i);
911 
912 		err = write_rss(pi, pi->rss);
913 		if (err)
914 			return err;
915 	}
916 	return 0;
917 }
918 
919 /*
920  * Return the channel of the ingress queue with the given qid.
921  */
922 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
923 {
924 	qid -= p->ingr_start;
925 	return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
926 }
927 
928 /*
929  * Wait until all NAPI handlers are descheduled.
930  */
931 static void quiesce_rx(struct adapter *adap)
932 {
933 	int i;
934 
935 	for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
936 		struct sge_rspq *q = adap->sge.ingr_map[i];
937 
938 		if (q && q->handler)
939 			napi_disable(&q->napi);
940 	}
941 }
942 
943 /*
944  * Enable NAPI scheduling and interrupt generation for all Rx queues.
945  */
946 static void enable_rx(struct adapter *adap)
947 {
948 	int i;
949 
950 	for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
951 		struct sge_rspq *q = adap->sge.ingr_map[i];
952 
953 		if (!q)
954 			continue;
955 		if (q->handler)
956 			napi_enable(&q->napi);
957 		/* 0-increment GTS to start the timer and enable interrupts */
958 		t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
959 			     SEINTARM(q->intr_params) |
960 			     INGRESSQID(q->cntxt_id));
961 	}
962 }
963 
964 /**
965  *	setup_sge_queues - configure SGE Tx/Rx/response queues
966  *	@adap: the adapter
967  *
968  *	Determines how many sets of SGE queues to use and initializes them.
969  *	We support multiple queue sets per port if we have MSI-X, otherwise
970  *	just one queue set per port.
971  */
972 static int setup_sge_queues(struct adapter *adap)
973 {
974 	int err, msi_idx, i, j;
975 	struct sge *s = &adap->sge;
976 
977 	bitmap_zero(s->starving_fl, MAX_EGRQ);
978 	bitmap_zero(s->txq_maperr, MAX_EGRQ);
979 
980 	if (adap->flags & USING_MSIX)
981 		msi_idx = 1;         /* vector 0 is for non-queue interrupts */
982 	else {
983 		err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
984 				       NULL, NULL);
985 		if (err)
986 			return err;
987 		msi_idx = -((int)s->intrq.abs_id + 1);
988 	}
989 
990 	err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
991 			       msi_idx, NULL, fwevtq_handler);
992 	if (err) {
993 freeout:	t4_free_sge_resources(adap);
994 		return err;
995 	}
996 
997 	for_each_port(adap, i) {
998 		struct net_device *dev = adap->port[i];
999 		struct port_info *pi = netdev_priv(dev);
1000 		struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1001 		struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1002 
1003 		for (j = 0; j < pi->nqsets; j++, q++) {
1004 			if (msi_idx > 0)
1005 				msi_idx++;
1006 			err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1007 					       msi_idx, &q->fl,
1008 					       t4_ethrx_handler);
1009 			if (err)
1010 				goto freeout;
1011 			q->rspq.idx = j;
1012 			memset(&q->stats, 0, sizeof(q->stats));
1013 		}
1014 		for (j = 0; j < pi->nqsets; j++, t++) {
1015 			err = t4_sge_alloc_eth_txq(adap, t, dev,
1016 					netdev_get_tx_queue(dev, j),
1017 					s->fw_evtq.cntxt_id);
1018 			if (err)
1019 				goto freeout;
1020 		}
1021 	}
1022 
1023 	j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
1024 	for_each_ofldrxq(s, i) {
1025 		struct sge_ofld_rxq *q = &s->ofldrxq[i];
1026 		struct net_device *dev = adap->port[i / j];
1027 
1028 		if (msi_idx > 0)
1029 			msi_idx++;
1030 		err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
1031 				       &q->fl, uldrx_handler);
1032 		if (err)
1033 			goto freeout;
1034 		memset(&q->stats, 0, sizeof(q->stats));
1035 		s->ofld_rxq[i] = q->rspq.abs_id;
1036 		err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
1037 					    s->fw_evtq.cntxt_id);
1038 		if (err)
1039 			goto freeout;
1040 	}
1041 
1042 	for_each_rdmarxq(s, i) {
1043 		struct sge_ofld_rxq *q = &s->rdmarxq[i];
1044 
1045 		if (msi_idx > 0)
1046 			msi_idx++;
1047 		err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
1048 				       msi_idx, &q->fl, uldrx_handler);
1049 		if (err)
1050 			goto freeout;
1051 		memset(&q->stats, 0, sizeof(q->stats));
1052 		s->rdma_rxq[i] = q->rspq.abs_id;
1053 	}
1054 
1055 	for_each_port(adap, i) {
1056 		/*
1057 		 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1058 		 * have RDMA queues, and that's the right value.
1059 		 */
1060 		err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1061 					    s->fw_evtq.cntxt_id,
1062 					    s->rdmarxq[i].rspq.cntxt_id);
1063 		if (err)
1064 			goto freeout;
1065 	}
1066 
1067 	t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1068 		     RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1069 		     QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1070 	return 0;
1071 }
1072 
1073 /*
1074  * Returns 0 if new FW was successfully loaded, a positive errno if a load was
1075  * started but failed, and a negative errno if flash load couldn't start.
1076  */
1077 static int upgrade_fw(struct adapter *adap)
1078 {
1079 	int ret;
1080 	u32 vers, exp_major;
1081 	const struct fw_hdr *hdr;
1082 	const struct firmware *fw;
1083 	struct device *dev = adap->pdev_dev;
1084 	char *fw_file_name;
1085 
1086 	switch (CHELSIO_CHIP_VERSION(adap->chip)) {
1087 	case CHELSIO_T4:
1088 		fw_file_name = FW_FNAME;
1089 		exp_major = FW_VERSION_MAJOR;
1090 		break;
1091 	case CHELSIO_T5:
1092 		fw_file_name = FW5_FNAME;
1093 		exp_major = FW_VERSION_MAJOR_T5;
1094 		break;
1095 	default:
1096 		dev_err(dev, "Unsupported chip type, %x\n", adap->chip);
1097 		return -EINVAL;
1098 	}
1099 
1100 	ret = request_firmware(&fw, fw_file_name, dev);
1101 	if (ret < 0) {
1102 		dev_err(dev, "unable to load firmware image %s, error %d\n",
1103 			fw_file_name, ret);
1104 		return ret;
1105 	}
1106 
1107 	hdr = (const struct fw_hdr *)fw->data;
1108 	vers = ntohl(hdr->fw_ver);
1109 	if (FW_HDR_FW_VER_MAJOR_GET(vers) != exp_major) {
1110 		ret = -EINVAL;              /* wrong major version, won't do */
1111 		goto out;
1112 	}
1113 
1114 	/*
1115 	 * If the flash FW is unusable or we found something newer, load it.
1116 	 */
1117 	if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != exp_major ||
1118 	    vers > adap->params.fw_vers) {
1119 		dev_info(dev, "upgrading firmware ...\n");
1120 		ret = t4_fw_upgrade(adap, adap->mbox, fw->data, fw->size,
1121 				    /*force=*/false);
1122 		if (!ret)
1123 			dev_info(dev,
1124 				 "firmware upgraded to version %pI4 from %s\n",
1125 				 &hdr->fw_ver, fw_file_name);
1126 		else
1127 			dev_err(dev, "firmware upgrade failed! err=%d\n", -ret);
1128 	} else {
1129 		/*
1130 		 * Tell our caller that we didn't upgrade the firmware.
1131 		 */
1132 		ret = -EINVAL;
1133 	}
1134 
1135 out:	release_firmware(fw);
1136 	return ret;
1137 }
1138 
1139 /*
1140  * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1141  * The allocated memory is cleared.
1142  */
1143 void *t4_alloc_mem(size_t size)
1144 {
1145 	void *p = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1146 
1147 	if (!p)
1148 		p = vzalloc(size);
1149 	return p;
1150 }
1151 
1152 /*
1153  * Free memory allocated through alloc_mem().
1154  */
1155 static void t4_free_mem(void *addr)
1156 {
1157 	if (is_vmalloc_addr(addr))
1158 		vfree(addr);
1159 	else
1160 		kfree(addr);
1161 }
1162 
1163 /* Send a Work Request to write the filter at a specified index.  We construct
1164  * a Firmware Filter Work Request to have the work done and put the indicated
1165  * filter into "pending" mode which will prevent any further actions against
1166  * it till we get a reply from the firmware on the completion status of the
1167  * request.
1168  */
1169 static int set_filter_wr(struct adapter *adapter, int fidx)
1170 {
1171 	struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1172 	struct sk_buff *skb;
1173 	struct fw_filter_wr *fwr;
1174 	unsigned int ftid;
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 			return -EAGAIN;
1185 		if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1186 					f->fs.eport, f->fs.dmac)) {
1187 			cxgb4_l2t_release(f->l2t);
1188 			f->l2t = NULL;
1189 			return -ENOMEM;
1190 		}
1191 	}
1192 
1193 	ftid = adapter->tids.ftid_base + fidx;
1194 
1195 	skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1196 	fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1197 	memset(fwr, 0, sizeof(*fwr));
1198 
1199 	/* It would be nice to put most of the following in t4_hw.c but most
1200 	 * of the work is translating the cxgbtool ch_filter_specification
1201 	 * into the Work Request and the definition of that structure is
1202 	 * currently in cxgbtool.h which isn't appropriate to pull into the
1203 	 * common code.  We may eventually try to come up with a more neutral
1204 	 * filter specification structure but for now it's easiest to simply
1205 	 * put this fairly direct code in line ...
1206 	 */
1207 	fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1208 	fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1209 	fwr->tid_to_iq =
1210 		htonl(V_FW_FILTER_WR_TID(ftid) |
1211 		      V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1212 		      V_FW_FILTER_WR_NOREPLY(0) |
1213 		      V_FW_FILTER_WR_IQ(f->fs.iq));
1214 	fwr->del_filter_to_l2tix =
1215 		htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1216 		      V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1217 		      V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1218 		      V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1219 		      V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1220 		      V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1221 		      V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1222 		      V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1223 		      V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1224 					     f->fs.newvlan == VLAN_REWRITE) |
1225 		      V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1226 					    f->fs.newvlan == VLAN_REWRITE) |
1227 		      V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1228 		      V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1229 		      V_FW_FILTER_WR_PRIO(f->fs.prio) |
1230 		      V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1231 	fwr->ethtype = htons(f->fs.val.ethtype);
1232 	fwr->ethtypem = htons(f->fs.mask.ethtype);
1233 	fwr->frag_to_ovlan_vldm =
1234 		(V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1235 		 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1236 		 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1237 		 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1238 		 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1239 		 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1240 	fwr->smac_sel = 0;
1241 	fwr->rx_chan_rx_rpl_iq =
1242 		htons(V_FW_FILTER_WR_RX_CHAN(0) |
1243 		      V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1244 	fwr->maci_to_matchtypem =
1245 		htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1246 		      V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1247 		      V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1248 		      V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1249 		      V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1250 		      V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1251 		      V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1252 		      V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1253 	fwr->ptcl = f->fs.val.proto;
1254 	fwr->ptclm = f->fs.mask.proto;
1255 	fwr->ttyp = f->fs.val.tos;
1256 	fwr->ttypm = f->fs.mask.tos;
1257 	fwr->ivlan = htons(f->fs.val.ivlan);
1258 	fwr->ivlanm = htons(f->fs.mask.ivlan);
1259 	fwr->ovlan = htons(f->fs.val.ovlan);
1260 	fwr->ovlanm = htons(f->fs.mask.ovlan);
1261 	memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1262 	memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1263 	memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1264 	memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1265 	fwr->lp = htons(f->fs.val.lport);
1266 	fwr->lpm = htons(f->fs.mask.lport);
1267 	fwr->fp = htons(f->fs.val.fport);
1268 	fwr->fpm = htons(f->fs.mask.fport);
1269 	if (f->fs.newsmac)
1270 		memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1271 
1272 	/* Mark the filter as "pending" and ship off the Filter Work Request.
1273 	 * When we get the Work Request Reply we'll clear the pending status.
1274 	 */
1275 	f->pending = 1;
1276 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1277 	t4_ofld_send(adapter, skb);
1278 	return 0;
1279 }
1280 
1281 /* Delete the filter at a specified index.
1282  */
1283 static int del_filter_wr(struct adapter *adapter, int fidx)
1284 {
1285 	struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1286 	struct sk_buff *skb;
1287 	struct fw_filter_wr *fwr;
1288 	unsigned int len, ftid;
1289 
1290 	len = sizeof(*fwr);
1291 	ftid = adapter->tids.ftid_base + fidx;
1292 
1293 	skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1294 	fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1295 	t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1296 
1297 	/* Mark the filter as "pending" and ship off the Filter Work Request.
1298 	 * When we get the Work Request Reply we'll clear the pending status.
1299 	 */
1300 	f->pending = 1;
1301 	t4_mgmt_tx(adapter, skb);
1302 	return 0;
1303 }
1304 
1305 static inline int is_offload(const struct adapter *adap)
1306 {
1307 	return adap->params.offload;
1308 }
1309 
1310 /*
1311  * Implementation of ethtool operations.
1312  */
1313 
1314 static u32 get_msglevel(struct net_device *dev)
1315 {
1316 	return netdev2adap(dev)->msg_enable;
1317 }
1318 
1319 static void set_msglevel(struct net_device *dev, u32 val)
1320 {
1321 	netdev2adap(dev)->msg_enable = val;
1322 }
1323 
1324 static char stats_strings[][ETH_GSTRING_LEN] = {
1325 	"TxOctetsOK         ",
1326 	"TxFramesOK         ",
1327 	"TxBroadcastFrames  ",
1328 	"TxMulticastFrames  ",
1329 	"TxUnicastFrames    ",
1330 	"TxErrorFrames      ",
1331 
1332 	"TxFrames64         ",
1333 	"TxFrames65To127    ",
1334 	"TxFrames128To255   ",
1335 	"TxFrames256To511   ",
1336 	"TxFrames512To1023  ",
1337 	"TxFrames1024To1518 ",
1338 	"TxFrames1519ToMax  ",
1339 
1340 	"TxFramesDropped    ",
1341 	"TxPauseFrames      ",
1342 	"TxPPP0Frames       ",
1343 	"TxPPP1Frames       ",
1344 	"TxPPP2Frames       ",
1345 	"TxPPP3Frames       ",
1346 	"TxPPP4Frames       ",
1347 	"TxPPP5Frames       ",
1348 	"TxPPP6Frames       ",
1349 	"TxPPP7Frames       ",
1350 
1351 	"RxOctetsOK         ",
1352 	"RxFramesOK         ",
1353 	"RxBroadcastFrames  ",
1354 	"RxMulticastFrames  ",
1355 	"RxUnicastFrames    ",
1356 
1357 	"RxFramesTooLong    ",
1358 	"RxJabberErrors     ",
1359 	"RxFCSErrors        ",
1360 	"RxLengthErrors     ",
1361 	"RxSymbolErrors     ",
1362 	"RxRuntFrames       ",
1363 
1364 	"RxFrames64         ",
1365 	"RxFrames65To127    ",
1366 	"RxFrames128To255   ",
1367 	"RxFrames256To511   ",
1368 	"RxFrames512To1023  ",
1369 	"RxFrames1024To1518 ",
1370 	"RxFrames1519ToMax  ",
1371 
1372 	"RxPauseFrames      ",
1373 	"RxPPP0Frames       ",
1374 	"RxPPP1Frames       ",
1375 	"RxPPP2Frames       ",
1376 	"RxPPP3Frames       ",
1377 	"RxPPP4Frames       ",
1378 	"RxPPP5Frames       ",
1379 	"RxPPP6Frames       ",
1380 	"RxPPP7Frames       ",
1381 
1382 	"RxBG0FramesDropped ",
1383 	"RxBG1FramesDropped ",
1384 	"RxBG2FramesDropped ",
1385 	"RxBG3FramesDropped ",
1386 	"RxBG0FramesTrunc   ",
1387 	"RxBG1FramesTrunc   ",
1388 	"RxBG2FramesTrunc   ",
1389 	"RxBG3FramesTrunc   ",
1390 
1391 	"TSO                ",
1392 	"TxCsumOffload      ",
1393 	"RxCsumGood         ",
1394 	"VLANextractions    ",
1395 	"VLANinsertions     ",
1396 	"GROpackets         ",
1397 	"GROmerged          ",
1398 	"WriteCoalSuccess   ",
1399 	"WriteCoalFail      ",
1400 };
1401 
1402 static int get_sset_count(struct net_device *dev, int sset)
1403 {
1404 	switch (sset) {
1405 	case ETH_SS_STATS:
1406 		return ARRAY_SIZE(stats_strings);
1407 	default:
1408 		return -EOPNOTSUPP;
1409 	}
1410 }
1411 
1412 #define T4_REGMAP_SIZE (160 * 1024)
1413 #define T5_REGMAP_SIZE (332 * 1024)
1414 
1415 static int get_regs_len(struct net_device *dev)
1416 {
1417 	struct adapter *adap = netdev2adap(dev);
1418 	if (is_t4(adap->chip))
1419 		return T4_REGMAP_SIZE;
1420 	else
1421 		return T5_REGMAP_SIZE;
1422 }
1423 
1424 static int get_eeprom_len(struct net_device *dev)
1425 {
1426 	return EEPROMSIZE;
1427 }
1428 
1429 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1430 {
1431 	struct adapter *adapter = netdev2adap(dev);
1432 
1433 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1434 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1435 	strlcpy(info->bus_info, pci_name(adapter->pdev),
1436 		sizeof(info->bus_info));
1437 
1438 	if (adapter->params.fw_vers)
1439 		snprintf(info->fw_version, sizeof(info->fw_version),
1440 			"%u.%u.%u.%u, TP %u.%u.%u.%u",
1441 			FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1442 			FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1443 			FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1444 			FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1445 			FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1446 			FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1447 			FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1448 			FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1449 }
1450 
1451 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1452 {
1453 	if (stringset == ETH_SS_STATS)
1454 		memcpy(data, stats_strings, sizeof(stats_strings));
1455 }
1456 
1457 /*
1458  * port stats maintained per queue of the port.  They should be in the same
1459  * order as in stats_strings above.
1460  */
1461 struct queue_port_stats {
1462 	u64 tso;
1463 	u64 tx_csum;
1464 	u64 rx_csum;
1465 	u64 vlan_ex;
1466 	u64 vlan_ins;
1467 	u64 gro_pkts;
1468 	u64 gro_merged;
1469 };
1470 
1471 static void collect_sge_port_stats(const struct adapter *adap,
1472 		const struct port_info *p, struct queue_port_stats *s)
1473 {
1474 	int i;
1475 	const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1476 	const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1477 
1478 	memset(s, 0, sizeof(*s));
1479 	for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1480 		s->tso += tx->tso;
1481 		s->tx_csum += tx->tx_cso;
1482 		s->rx_csum += rx->stats.rx_cso;
1483 		s->vlan_ex += rx->stats.vlan_ex;
1484 		s->vlan_ins += tx->vlan_ins;
1485 		s->gro_pkts += rx->stats.lro_pkts;
1486 		s->gro_merged += rx->stats.lro_merged;
1487 	}
1488 }
1489 
1490 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1491 		      u64 *data)
1492 {
1493 	struct port_info *pi = netdev_priv(dev);
1494 	struct adapter *adapter = pi->adapter;
1495 	u32 val1, val2;
1496 
1497 	t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1498 
1499 	data += sizeof(struct port_stats) / sizeof(u64);
1500 	collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1501 	data += sizeof(struct queue_port_stats) / sizeof(u64);
1502 	if (!is_t4(adapter->chip)) {
1503 		t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1504 		val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1505 		val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1506 		*data = val1 - val2;
1507 		data++;
1508 		*data = val2;
1509 		data++;
1510 	} else {
1511 		memset(data, 0, 2 * sizeof(u64));
1512 		*data += 2;
1513 	}
1514 }
1515 
1516 /*
1517  * Return a version number to identify the type of adapter.  The scheme is:
1518  * - bits 0..9: chip version
1519  * - bits 10..15: chip revision
1520  * - bits 16..23: register dump version
1521  */
1522 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1523 {
1524 	return CHELSIO_CHIP_VERSION(ap->chip) |
1525 		(CHELSIO_CHIP_RELEASE(ap->chip) << 10) | (1 << 16);
1526 }
1527 
1528 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1529 			   unsigned int end)
1530 {
1531 	u32 *p = buf + start;
1532 
1533 	for ( ; start <= end; start += sizeof(u32))
1534 		*p++ = t4_read_reg(ap, start);
1535 }
1536 
1537 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1538 		     void *buf)
1539 {
1540 	static const unsigned int t4_reg_ranges[] = {
1541 		0x1008, 0x1108,
1542 		0x1180, 0x11b4,
1543 		0x11fc, 0x123c,
1544 		0x1300, 0x173c,
1545 		0x1800, 0x18fc,
1546 		0x3000, 0x30d8,
1547 		0x30e0, 0x5924,
1548 		0x5960, 0x59d4,
1549 		0x5a00, 0x5af8,
1550 		0x6000, 0x6098,
1551 		0x6100, 0x6150,
1552 		0x6200, 0x6208,
1553 		0x6240, 0x6248,
1554 		0x6280, 0x6338,
1555 		0x6370, 0x638c,
1556 		0x6400, 0x643c,
1557 		0x6500, 0x6524,
1558 		0x6a00, 0x6a38,
1559 		0x6a60, 0x6a78,
1560 		0x6b00, 0x6b84,
1561 		0x6bf0, 0x6c84,
1562 		0x6cf0, 0x6d84,
1563 		0x6df0, 0x6e84,
1564 		0x6ef0, 0x6f84,
1565 		0x6ff0, 0x7084,
1566 		0x70f0, 0x7184,
1567 		0x71f0, 0x7284,
1568 		0x72f0, 0x7384,
1569 		0x73f0, 0x7450,
1570 		0x7500, 0x7530,
1571 		0x7600, 0x761c,
1572 		0x7680, 0x76cc,
1573 		0x7700, 0x7798,
1574 		0x77c0, 0x77fc,
1575 		0x7900, 0x79fc,
1576 		0x7b00, 0x7c38,
1577 		0x7d00, 0x7efc,
1578 		0x8dc0, 0x8e1c,
1579 		0x8e30, 0x8e78,
1580 		0x8ea0, 0x8f6c,
1581 		0x8fc0, 0x9074,
1582 		0x90fc, 0x90fc,
1583 		0x9400, 0x9458,
1584 		0x9600, 0x96bc,
1585 		0x9800, 0x9808,
1586 		0x9820, 0x983c,
1587 		0x9850, 0x9864,
1588 		0x9c00, 0x9c6c,
1589 		0x9c80, 0x9cec,
1590 		0x9d00, 0x9d6c,
1591 		0x9d80, 0x9dec,
1592 		0x9e00, 0x9e6c,
1593 		0x9e80, 0x9eec,
1594 		0x9f00, 0x9f6c,
1595 		0x9f80, 0x9fec,
1596 		0xd004, 0xd03c,
1597 		0xdfc0, 0xdfe0,
1598 		0xe000, 0xea7c,
1599 		0xf000, 0x11190,
1600 		0x19040, 0x1906c,
1601 		0x19078, 0x19080,
1602 		0x1908c, 0x19124,
1603 		0x19150, 0x191b0,
1604 		0x191d0, 0x191e8,
1605 		0x19238, 0x1924c,
1606 		0x193f8, 0x19474,
1607 		0x19490, 0x194f8,
1608 		0x19800, 0x19f30,
1609 		0x1a000, 0x1a06c,
1610 		0x1a0b0, 0x1a120,
1611 		0x1a128, 0x1a138,
1612 		0x1a190, 0x1a1c4,
1613 		0x1a1fc, 0x1a1fc,
1614 		0x1e040, 0x1e04c,
1615 		0x1e284, 0x1e28c,
1616 		0x1e2c0, 0x1e2c0,
1617 		0x1e2e0, 0x1e2e0,
1618 		0x1e300, 0x1e384,
1619 		0x1e3c0, 0x1e3c8,
1620 		0x1e440, 0x1e44c,
1621 		0x1e684, 0x1e68c,
1622 		0x1e6c0, 0x1e6c0,
1623 		0x1e6e0, 0x1e6e0,
1624 		0x1e700, 0x1e784,
1625 		0x1e7c0, 0x1e7c8,
1626 		0x1e840, 0x1e84c,
1627 		0x1ea84, 0x1ea8c,
1628 		0x1eac0, 0x1eac0,
1629 		0x1eae0, 0x1eae0,
1630 		0x1eb00, 0x1eb84,
1631 		0x1ebc0, 0x1ebc8,
1632 		0x1ec40, 0x1ec4c,
1633 		0x1ee84, 0x1ee8c,
1634 		0x1eec0, 0x1eec0,
1635 		0x1eee0, 0x1eee0,
1636 		0x1ef00, 0x1ef84,
1637 		0x1efc0, 0x1efc8,
1638 		0x1f040, 0x1f04c,
1639 		0x1f284, 0x1f28c,
1640 		0x1f2c0, 0x1f2c0,
1641 		0x1f2e0, 0x1f2e0,
1642 		0x1f300, 0x1f384,
1643 		0x1f3c0, 0x1f3c8,
1644 		0x1f440, 0x1f44c,
1645 		0x1f684, 0x1f68c,
1646 		0x1f6c0, 0x1f6c0,
1647 		0x1f6e0, 0x1f6e0,
1648 		0x1f700, 0x1f784,
1649 		0x1f7c0, 0x1f7c8,
1650 		0x1f840, 0x1f84c,
1651 		0x1fa84, 0x1fa8c,
1652 		0x1fac0, 0x1fac0,
1653 		0x1fae0, 0x1fae0,
1654 		0x1fb00, 0x1fb84,
1655 		0x1fbc0, 0x1fbc8,
1656 		0x1fc40, 0x1fc4c,
1657 		0x1fe84, 0x1fe8c,
1658 		0x1fec0, 0x1fec0,
1659 		0x1fee0, 0x1fee0,
1660 		0x1ff00, 0x1ff84,
1661 		0x1ffc0, 0x1ffc8,
1662 		0x20000, 0x2002c,
1663 		0x20100, 0x2013c,
1664 		0x20190, 0x201c8,
1665 		0x20200, 0x20318,
1666 		0x20400, 0x20528,
1667 		0x20540, 0x20614,
1668 		0x21000, 0x21040,
1669 		0x2104c, 0x21060,
1670 		0x210c0, 0x210ec,
1671 		0x21200, 0x21268,
1672 		0x21270, 0x21284,
1673 		0x212fc, 0x21388,
1674 		0x21400, 0x21404,
1675 		0x21500, 0x21518,
1676 		0x2152c, 0x2153c,
1677 		0x21550, 0x21554,
1678 		0x21600, 0x21600,
1679 		0x21608, 0x21628,
1680 		0x21630, 0x2163c,
1681 		0x21700, 0x2171c,
1682 		0x21780, 0x2178c,
1683 		0x21800, 0x21c38,
1684 		0x21c80, 0x21d7c,
1685 		0x21e00, 0x21e04,
1686 		0x22000, 0x2202c,
1687 		0x22100, 0x2213c,
1688 		0x22190, 0x221c8,
1689 		0x22200, 0x22318,
1690 		0x22400, 0x22528,
1691 		0x22540, 0x22614,
1692 		0x23000, 0x23040,
1693 		0x2304c, 0x23060,
1694 		0x230c0, 0x230ec,
1695 		0x23200, 0x23268,
1696 		0x23270, 0x23284,
1697 		0x232fc, 0x23388,
1698 		0x23400, 0x23404,
1699 		0x23500, 0x23518,
1700 		0x2352c, 0x2353c,
1701 		0x23550, 0x23554,
1702 		0x23600, 0x23600,
1703 		0x23608, 0x23628,
1704 		0x23630, 0x2363c,
1705 		0x23700, 0x2371c,
1706 		0x23780, 0x2378c,
1707 		0x23800, 0x23c38,
1708 		0x23c80, 0x23d7c,
1709 		0x23e00, 0x23e04,
1710 		0x24000, 0x2402c,
1711 		0x24100, 0x2413c,
1712 		0x24190, 0x241c8,
1713 		0x24200, 0x24318,
1714 		0x24400, 0x24528,
1715 		0x24540, 0x24614,
1716 		0x25000, 0x25040,
1717 		0x2504c, 0x25060,
1718 		0x250c0, 0x250ec,
1719 		0x25200, 0x25268,
1720 		0x25270, 0x25284,
1721 		0x252fc, 0x25388,
1722 		0x25400, 0x25404,
1723 		0x25500, 0x25518,
1724 		0x2552c, 0x2553c,
1725 		0x25550, 0x25554,
1726 		0x25600, 0x25600,
1727 		0x25608, 0x25628,
1728 		0x25630, 0x2563c,
1729 		0x25700, 0x2571c,
1730 		0x25780, 0x2578c,
1731 		0x25800, 0x25c38,
1732 		0x25c80, 0x25d7c,
1733 		0x25e00, 0x25e04,
1734 		0x26000, 0x2602c,
1735 		0x26100, 0x2613c,
1736 		0x26190, 0x261c8,
1737 		0x26200, 0x26318,
1738 		0x26400, 0x26528,
1739 		0x26540, 0x26614,
1740 		0x27000, 0x27040,
1741 		0x2704c, 0x27060,
1742 		0x270c0, 0x270ec,
1743 		0x27200, 0x27268,
1744 		0x27270, 0x27284,
1745 		0x272fc, 0x27388,
1746 		0x27400, 0x27404,
1747 		0x27500, 0x27518,
1748 		0x2752c, 0x2753c,
1749 		0x27550, 0x27554,
1750 		0x27600, 0x27600,
1751 		0x27608, 0x27628,
1752 		0x27630, 0x2763c,
1753 		0x27700, 0x2771c,
1754 		0x27780, 0x2778c,
1755 		0x27800, 0x27c38,
1756 		0x27c80, 0x27d7c,
1757 		0x27e00, 0x27e04
1758 	};
1759 
1760 	static const unsigned int t5_reg_ranges[] = {
1761 		0x1008, 0x1148,
1762 		0x1180, 0x11b4,
1763 		0x11fc, 0x123c,
1764 		0x1280, 0x173c,
1765 		0x1800, 0x18fc,
1766 		0x3000, 0x3028,
1767 		0x3060, 0x30d8,
1768 		0x30e0, 0x30fc,
1769 		0x3140, 0x357c,
1770 		0x35a8, 0x35cc,
1771 		0x35ec, 0x35ec,
1772 		0x3600, 0x5624,
1773 		0x56cc, 0x575c,
1774 		0x580c, 0x5814,
1775 		0x5890, 0x58bc,
1776 		0x5940, 0x59dc,
1777 		0x59fc, 0x5a18,
1778 		0x5a60, 0x5a9c,
1779 		0x5b9c, 0x5bfc,
1780 		0x6000, 0x6040,
1781 		0x6058, 0x614c,
1782 		0x7700, 0x7798,
1783 		0x77c0, 0x78fc,
1784 		0x7b00, 0x7c54,
1785 		0x7d00, 0x7efc,
1786 		0x8dc0, 0x8de0,
1787 		0x8df8, 0x8e84,
1788 		0x8ea0, 0x8f84,
1789 		0x8fc0, 0x90f8,
1790 		0x9400, 0x9470,
1791 		0x9600, 0x96f4,
1792 		0x9800, 0x9808,
1793 		0x9820, 0x983c,
1794 		0x9850, 0x9864,
1795 		0x9c00, 0x9c6c,
1796 		0x9c80, 0x9cec,
1797 		0x9d00, 0x9d6c,
1798 		0x9d80, 0x9dec,
1799 		0x9e00, 0x9e6c,
1800 		0x9e80, 0x9eec,
1801 		0x9f00, 0x9f6c,
1802 		0x9f80, 0xa020,
1803 		0xd004, 0xd03c,
1804 		0xdfc0, 0xdfe0,
1805 		0xe000, 0x11088,
1806 		0x1109c, 0x1117c,
1807 		0x11190, 0x11204,
1808 		0x19040, 0x1906c,
1809 		0x19078, 0x19080,
1810 		0x1908c, 0x19124,
1811 		0x19150, 0x191b0,
1812 		0x191d0, 0x191e8,
1813 		0x19238, 0x19290,
1814 		0x193f8, 0x19474,
1815 		0x19490, 0x194cc,
1816 		0x194f0, 0x194f8,
1817 		0x19c00, 0x19c60,
1818 		0x19c94, 0x19e10,
1819 		0x19e50, 0x19f34,
1820 		0x19f40, 0x19f50,
1821 		0x19f90, 0x19fe4,
1822 		0x1a000, 0x1a06c,
1823 		0x1a0b0, 0x1a120,
1824 		0x1a128, 0x1a138,
1825 		0x1a190, 0x1a1c4,
1826 		0x1a1fc, 0x1a1fc,
1827 		0x1e008, 0x1e00c,
1828 		0x1e040, 0x1e04c,
1829 		0x1e284, 0x1e290,
1830 		0x1e2c0, 0x1e2c0,
1831 		0x1e2e0, 0x1e2e0,
1832 		0x1e300, 0x1e384,
1833 		0x1e3c0, 0x1e3c8,
1834 		0x1e408, 0x1e40c,
1835 		0x1e440, 0x1e44c,
1836 		0x1e684, 0x1e690,
1837 		0x1e6c0, 0x1e6c0,
1838 		0x1e6e0, 0x1e6e0,
1839 		0x1e700, 0x1e784,
1840 		0x1e7c0, 0x1e7c8,
1841 		0x1e808, 0x1e80c,
1842 		0x1e840, 0x1e84c,
1843 		0x1ea84, 0x1ea90,
1844 		0x1eac0, 0x1eac0,
1845 		0x1eae0, 0x1eae0,
1846 		0x1eb00, 0x1eb84,
1847 		0x1ebc0, 0x1ebc8,
1848 		0x1ec08, 0x1ec0c,
1849 		0x1ec40, 0x1ec4c,
1850 		0x1ee84, 0x1ee90,
1851 		0x1eec0, 0x1eec0,
1852 		0x1eee0, 0x1eee0,
1853 		0x1ef00, 0x1ef84,
1854 		0x1efc0, 0x1efc8,
1855 		0x1f008, 0x1f00c,
1856 		0x1f040, 0x1f04c,
1857 		0x1f284, 0x1f290,
1858 		0x1f2c0, 0x1f2c0,
1859 		0x1f2e0, 0x1f2e0,
1860 		0x1f300, 0x1f384,
1861 		0x1f3c0, 0x1f3c8,
1862 		0x1f408, 0x1f40c,
1863 		0x1f440, 0x1f44c,
1864 		0x1f684, 0x1f690,
1865 		0x1f6c0, 0x1f6c0,
1866 		0x1f6e0, 0x1f6e0,
1867 		0x1f700, 0x1f784,
1868 		0x1f7c0, 0x1f7c8,
1869 		0x1f808, 0x1f80c,
1870 		0x1f840, 0x1f84c,
1871 		0x1fa84, 0x1fa90,
1872 		0x1fac0, 0x1fac0,
1873 		0x1fae0, 0x1fae0,
1874 		0x1fb00, 0x1fb84,
1875 		0x1fbc0, 0x1fbc8,
1876 		0x1fc08, 0x1fc0c,
1877 		0x1fc40, 0x1fc4c,
1878 		0x1fe84, 0x1fe90,
1879 		0x1fec0, 0x1fec0,
1880 		0x1fee0, 0x1fee0,
1881 		0x1ff00, 0x1ff84,
1882 		0x1ffc0, 0x1ffc8,
1883 		0x30000, 0x30030,
1884 		0x30100, 0x30144,
1885 		0x30190, 0x301d0,
1886 		0x30200, 0x30318,
1887 		0x30400, 0x3052c,
1888 		0x30540, 0x3061c,
1889 		0x30800, 0x30834,
1890 		0x308c0, 0x30908,
1891 		0x30910, 0x309ac,
1892 		0x30a00, 0x30a04,
1893 		0x30a0c, 0x30a2c,
1894 		0x30a44, 0x30a50,
1895 		0x30a74, 0x30c24,
1896 		0x30d08, 0x30d14,
1897 		0x30d1c, 0x30d20,
1898 		0x30d3c, 0x30d50,
1899 		0x31200, 0x3120c,
1900 		0x31220, 0x31220,
1901 		0x31240, 0x31240,
1902 		0x31600, 0x31600,
1903 		0x31608, 0x3160c,
1904 		0x31a00, 0x31a1c,
1905 		0x31e04, 0x31e20,
1906 		0x31e38, 0x31e3c,
1907 		0x31e80, 0x31e80,
1908 		0x31e88, 0x31ea8,
1909 		0x31eb0, 0x31eb4,
1910 		0x31ec8, 0x31ed4,
1911 		0x31fb8, 0x32004,
1912 		0x32208, 0x3223c,
1913 		0x32600, 0x32630,
1914 		0x32a00, 0x32abc,
1915 		0x32b00, 0x32b70,
1916 		0x33000, 0x33048,
1917 		0x33060, 0x3309c,
1918 		0x330f0, 0x33148,
1919 		0x33160, 0x3319c,
1920 		0x331f0, 0x332e4,
1921 		0x332f8, 0x333e4,
1922 		0x333f8, 0x33448,
1923 		0x33460, 0x3349c,
1924 		0x334f0, 0x33548,
1925 		0x33560, 0x3359c,
1926 		0x335f0, 0x336e4,
1927 		0x336f8, 0x337e4,
1928 		0x337f8, 0x337fc,
1929 		0x33814, 0x33814,
1930 		0x3382c, 0x3382c,
1931 		0x33880, 0x3388c,
1932 		0x338e8, 0x338ec,
1933 		0x33900, 0x33948,
1934 		0x33960, 0x3399c,
1935 		0x339f0, 0x33ae4,
1936 		0x33af8, 0x33b10,
1937 		0x33b28, 0x33b28,
1938 		0x33b3c, 0x33b50,
1939 		0x33bf0, 0x33c10,
1940 		0x33c28, 0x33c28,
1941 		0x33c3c, 0x33c50,
1942 		0x33cf0, 0x33cfc,
1943 		0x34000, 0x34030,
1944 		0x34100, 0x34144,
1945 		0x34190, 0x341d0,
1946 		0x34200, 0x34318,
1947 		0x34400, 0x3452c,
1948 		0x34540, 0x3461c,
1949 		0x34800, 0x34834,
1950 		0x348c0, 0x34908,
1951 		0x34910, 0x349ac,
1952 		0x34a00, 0x34a04,
1953 		0x34a0c, 0x34a2c,
1954 		0x34a44, 0x34a50,
1955 		0x34a74, 0x34c24,
1956 		0x34d08, 0x34d14,
1957 		0x34d1c, 0x34d20,
1958 		0x34d3c, 0x34d50,
1959 		0x35200, 0x3520c,
1960 		0x35220, 0x35220,
1961 		0x35240, 0x35240,
1962 		0x35600, 0x35600,
1963 		0x35608, 0x3560c,
1964 		0x35a00, 0x35a1c,
1965 		0x35e04, 0x35e20,
1966 		0x35e38, 0x35e3c,
1967 		0x35e80, 0x35e80,
1968 		0x35e88, 0x35ea8,
1969 		0x35eb0, 0x35eb4,
1970 		0x35ec8, 0x35ed4,
1971 		0x35fb8, 0x36004,
1972 		0x36208, 0x3623c,
1973 		0x36600, 0x36630,
1974 		0x36a00, 0x36abc,
1975 		0x36b00, 0x36b70,
1976 		0x37000, 0x37048,
1977 		0x37060, 0x3709c,
1978 		0x370f0, 0x37148,
1979 		0x37160, 0x3719c,
1980 		0x371f0, 0x372e4,
1981 		0x372f8, 0x373e4,
1982 		0x373f8, 0x37448,
1983 		0x37460, 0x3749c,
1984 		0x374f0, 0x37548,
1985 		0x37560, 0x3759c,
1986 		0x375f0, 0x376e4,
1987 		0x376f8, 0x377e4,
1988 		0x377f8, 0x377fc,
1989 		0x37814, 0x37814,
1990 		0x3782c, 0x3782c,
1991 		0x37880, 0x3788c,
1992 		0x378e8, 0x378ec,
1993 		0x37900, 0x37948,
1994 		0x37960, 0x3799c,
1995 		0x379f0, 0x37ae4,
1996 		0x37af8, 0x37b10,
1997 		0x37b28, 0x37b28,
1998 		0x37b3c, 0x37b50,
1999 		0x37bf0, 0x37c10,
2000 		0x37c28, 0x37c28,
2001 		0x37c3c, 0x37c50,
2002 		0x37cf0, 0x37cfc,
2003 		0x38000, 0x38030,
2004 		0x38100, 0x38144,
2005 		0x38190, 0x381d0,
2006 		0x38200, 0x38318,
2007 		0x38400, 0x3852c,
2008 		0x38540, 0x3861c,
2009 		0x38800, 0x38834,
2010 		0x388c0, 0x38908,
2011 		0x38910, 0x389ac,
2012 		0x38a00, 0x38a04,
2013 		0x38a0c, 0x38a2c,
2014 		0x38a44, 0x38a50,
2015 		0x38a74, 0x38c24,
2016 		0x38d08, 0x38d14,
2017 		0x38d1c, 0x38d20,
2018 		0x38d3c, 0x38d50,
2019 		0x39200, 0x3920c,
2020 		0x39220, 0x39220,
2021 		0x39240, 0x39240,
2022 		0x39600, 0x39600,
2023 		0x39608, 0x3960c,
2024 		0x39a00, 0x39a1c,
2025 		0x39e04, 0x39e20,
2026 		0x39e38, 0x39e3c,
2027 		0x39e80, 0x39e80,
2028 		0x39e88, 0x39ea8,
2029 		0x39eb0, 0x39eb4,
2030 		0x39ec8, 0x39ed4,
2031 		0x39fb8, 0x3a004,
2032 		0x3a208, 0x3a23c,
2033 		0x3a600, 0x3a630,
2034 		0x3aa00, 0x3aabc,
2035 		0x3ab00, 0x3ab70,
2036 		0x3b000, 0x3b048,
2037 		0x3b060, 0x3b09c,
2038 		0x3b0f0, 0x3b148,
2039 		0x3b160, 0x3b19c,
2040 		0x3b1f0, 0x3b2e4,
2041 		0x3b2f8, 0x3b3e4,
2042 		0x3b3f8, 0x3b448,
2043 		0x3b460, 0x3b49c,
2044 		0x3b4f0, 0x3b548,
2045 		0x3b560, 0x3b59c,
2046 		0x3b5f0, 0x3b6e4,
2047 		0x3b6f8, 0x3b7e4,
2048 		0x3b7f8, 0x3b7fc,
2049 		0x3b814, 0x3b814,
2050 		0x3b82c, 0x3b82c,
2051 		0x3b880, 0x3b88c,
2052 		0x3b8e8, 0x3b8ec,
2053 		0x3b900, 0x3b948,
2054 		0x3b960, 0x3b99c,
2055 		0x3b9f0, 0x3bae4,
2056 		0x3baf8, 0x3bb10,
2057 		0x3bb28, 0x3bb28,
2058 		0x3bb3c, 0x3bb50,
2059 		0x3bbf0, 0x3bc10,
2060 		0x3bc28, 0x3bc28,
2061 		0x3bc3c, 0x3bc50,
2062 		0x3bcf0, 0x3bcfc,
2063 		0x3c000, 0x3c030,
2064 		0x3c100, 0x3c144,
2065 		0x3c190, 0x3c1d0,
2066 		0x3c200, 0x3c318,
2067 		0x3c400, 0x3c52c,
2068 		0x3c540, 0x3c61c,
2069 		0x3c800, 0x3c834,
2070 		0x3c8c0, 0x3c908,
2071 		0x3c910, 0x3c9ac,
2072 		0x3ca00, 0x3ca04,
2073 		0x3ca0c, 0x3ca2c,
2074 		0x3ca44, 0x3ca50,
2075 		0x3ca74, 0x3cc24,
2076 		0x3cd08, 0x3cd14,
2077 		0x3cd1c, 0x3cd20,
2078 		0x3cd3c, 0x3cd50,
2079 		0x3d200, 0x3d20c,
2080 		0x3d220, 0x3d220,
2081 		0x3d240, 0x3d240,
2082 		0x3d600, 0x3d600,
2083 		0x3d608, 0x3d60c,
2084 		0x3da00, 0x3da1c,
2085 		0x3de04, 0x3de20,
2086 		0x3de38, 0x3de3c,
2087 		0x3de80, 0x3de80,
2088 		0x3de88, 0x3dea8,
2089 		0x3deb0, 0x3deb4,
2090 		0x3dec8, 0x3ded4,
2091 		0x3dfb8, 0x3e004,
2092 		0x3e208, 0x3e23c,
2093 		0x3e600, 0x3e630,
2094 		0x3ea00, 0x3eabc,
2095 		0x3eb00, 0x3eb70,
2096 		0x3f000, 0x3f048,
2097 		0x3f060, 0x3f09c,
2098 		0x3f0f0, 0x3f148,
2099 		0x3f160, 0x3f19c,
2100 		0x3f1f0, 0x3f2e4,
2101 		0x3f2f8, 0x3f3e4,
2102 		0x3f3f8, 0x3f448,
2103 		0x3f460, 0x3f49c,
2104 		0x3f4f0, 0x3f548,
2105 		0x3f560, 0x3f59c,
2106 		0x3f5f0, 0x3f6e4,
2107 		0x3f6f8, 0x3f7e4,
2108 		0x3f7f8, 0x3f7fc,
2109 		0x3f814, 0x3f814,
2110 		0x3f82c, 0x3f82c,
2111 		0x3f880, 0x3f88c,
2112 		0x3f8e8, 0x3f8ec,
2113 		0x3f900, 0x3f948,
2114 		0x3f960, 0x3f99c,
2115 		0x3f9f0, 0x3fae4,
2116 		0x3faf8, 0x3fb10,
2117 		0x3fb28, 0x3fb28,
2118 		0x3fb3c, 0x3fb50,
2119 		0x3fbf0, 0x3fc10,
2120 		0x3fc28, 0x3fc28,
2121 		0x3fc3c, 0x3fc50,
2122 		0x3fcf0, 0x3fcfc,
2123 		0x40000, 0x4000c,
2124 		0x40040, 0x40068,
2125 		0x40080, 0x40144,
2126 		0x40180, 0x4018c,
2127 		0x40200, 0x40298,
2128 		0x402ac, 0x4033c,
2129 		0x403f8, 0x403fc,
2130 		0x41300, 0x413c4,
2131 		0x41400, 0x4141c,
2132 		0x41480, 0x414d0,
2133 		0x44000, 0x44078,
2134 		0x440c0, 0x44278,
2135 		0x442c0, 0x44478,
2136 		0x444c0, 0x44678,
2137 		0x446c0, 0x44878,
2138 		0x448c0, 0x449fc,
2139 		0x45000, 0x45068,
2140 		0x45080, 0x45084,
2141 		0x450a0, 0x450b0,
2142 		0x45200, 0x45268,
2143 		0x45280, 0x45284,
2144 		0x452a0, 0x452b0,
2145 		0x460c0, 0x460e4,
2146 		0x47000, 0x4708c,
2147 		0x47200, 0x47250,
2148 		0x47400, 0x47420,
2149 		0x47600, 0x47618,
2150 		0x47800, 0x47814,
2151 		0x48000, 0x4800c,
2152 		0x48040, 0x48068,
2153 		0x48080, 0x48144,
2154 		0x48180, 0x4818c,
2155 		0x48200, 0x48298,
2156 		0x482ac, 0x4833c,
2157 		0x483f8, 0x483fc,
2158 		0x49300, 0x493c4,
2159 		0x49400, 0x4941c,
2160 		0x49480, 0x494d0,
2161 		0x4c000, 0x4c078,
2162 		0x4c0c0, 0x4c278,
2163 		0x4c2c0, 0x4c478,
2164 		0x4c4c0, 0x4c678,
2165 		0x4c6c0, 0x4c878,
2166 		0x4c8c0, 0x4c9fc,
2167 		0x4d000, 0x4d068,
2168 		0x4d080, 0x4d084,
2169 		0x4d0a0, 0x4d0b0,
2170 		0x4d200, 0x4d268,
2171 		0x4d280, 0x4d284,
2172 		0x4d2a0, 0x4d2b0,
2173 		0x4e0c0, 0x4e0e4,
2174 		0x4f000, 0x4f08c,
2175 		0x4f200, 0x4f250,
2176 		0x4f400, 0x4f420,
2177 		0x4f600, 0x4f618,
2178 		0x4f800, 0x4f814,
2179 		0x50000, 0x500cc,
2180 		0x50400, 0x50400,
2181 		0x50800, 0x508cc,
2182 		0x50c00, 0x50c00,
2183 		0x51000, 0x5101c,
2184 		0x51300, 0x51308,
2185 	};
2186 
2187 	int i;
2188 	struct adapter *ap = netdev2adap(dev);
2189 	static const unsigned int *reg_ranges;
2190 	int arr_size = 0, buf_size = 0;
2191 
2192 	if (is_t4(ap->chip)) {
2193 		reg_ranges = &t4_reg_ranges[0];
2194 		arr_size = ARRAY_SIZE(t4_reg_ranges);
2195 		buf_size = T4_REGMAP_SIZE;
2196 	} else {
2197 		reg_ranges = &t5_reg_ranges[0];
2198 		arr_size = ARRAY_SIZE(t5_reg_ranges);
2199 		buf_size = T5_REGMAP_SIZE;
2200 	}
2201 
2202 	regs->version = mk_adap_vers(ap);
2203 
2204 	memset(buf, 0, buf_size);
2205 	for (i = 0; i < arr_size; i += 2)
2206 		reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2207 }
2208 
2209 static int restart_autoneg(struct net_device *dev)
2210 {
2211 	struct port_info *p = netdev_priv(dev);
2212 
2213 	if (!netif_running(dev))
2214 		return -EAGAIN;
2215 	if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2216 		return -EINVAL;
2217 	t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2218 	return 0;
2219 }
2220 
2221 static int identify_port(struct net_device *dev,
2222 			 enum ethtool_phys_id_state state)
2223 {
2224 	unsigned int val;
2225 	struct adapter *adap = netdev2adap(dev);
2226 
2227 	if (state == ETHTOOL_ID_ACTIVE)
2228 		val = 0xffff;
2229 	else if (state == ETHTOOL_ID_INACTIVE)
2230 		val = 0;
2231 	else
2232 		return -EINVAL;
2233 
2234 	return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2235 }
2236 
2237 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2238 {
2239 	unsigned int v = 0;
2240 
2241 	if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2242 	    type == FW_PORT_TYPE_BT_XAUI) {
2243 		v |= SUPPORTED_TP;
2244 		if (caps & FW_PORT_CAP_SPEED_100M)
2245 			v |= SUPPORTED_100baseT_Full;
2246 		if (caps & FW_PORT_CAP_SPEED_1G)
2247 			v |= SUPPORTED_1000baseT_Full;
2248 		if (caps & FW_PORT_CAP_SPEED_10G)
2249 			v |= SUPPORTED_10000baseT_Full;
2250 	} else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2251 		v |= SUPPORTED_Backplane;
2252 		if (caps & FW_PORT_CAP_SPEED_1G)
2253 			v |= SUPPORTED_1000baseKX_Full;
2254 		if (caps & FW_PORT_CAP_SPEED_10G)
2255 			v |= SUPPORTED_10000baseKX4_Full;
2256 	} else if (type == FW_PORT_TYPE_KR)
2257 		v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2258 	else if (type == FW_PORT_TYPE_BP_AP)
2259 		v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2260 		     SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2261 	else if (type == FW_PORT_TYPE_BP4_AP)
2262 		v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2263 		     SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2264 		     SUPPORTED_10000baseKX4_Full;
2265 	else if (type == FW_PORT_TYPE_FIBER_XFI ||
2266 		 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2267 		v |= SUPPORTED_FIBRE;
2268 
2269 	if (caps & FW_PORT_CAP_ANEG)
2270 		v |= SUPPORTED_Autoneg;
2271 	return v;
2272 }
2273 
2274 static unsigned int to_fw_linkcaps(unsigned int caps)
2275 {
2276 	unsigned int v = 0;
2277 
2278 	if (caps & ADVERTISED_100baseT_Full)
2279 		v |= FW_PORT_CAP_SPEED_100M;
2280 	if (caps & ADVERTISED_1000baseT_Full)
2281 		v |= FW_PORT_CAP_SPEED_1G;
2282 	if (caps & ADVERTISED_10000baseT_Full)
2283 		v |= FW_PORT_CAP_SPEED_10G;
2284 	return v;
2285 }
2286 
2287 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2288 {
2289 	const struct port_info *p = netdev_priv(dev);
2290 
2291 	if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2292 	    p->port_type == FW_PORT_TYPE_BT_XFI ||
2293 	    p->port_type == FW_PORT_TYPE_BT_XAUI)
2294 		cmd->port = PORT_TP;
2295 	else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2296 		 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2297 		cmd->port = PORT_FIBRE;
2298 	else if (p->port_type == FW_PORT_TYPE_SFP) {
2299 		if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2300 		    p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2301 			cmd->port = PORT_DA;
2302 		else
2303 			cmd->port = PORT_FIBRE;
2304 	} else
2305 		cmd->port = PORT_OTHER;
2306 
2307 	if (p->mdio_addr >= 0) {
2308 		cmd->phy_address = p->mdio_addr;
2309 		cmd->transceiver = XCVR_EXTERNAL;
2310 		cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2311 			MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2312 	} else {
2313 		cmd->phy_address = 0;  /* not really, but no better option */
2314 		cmd->transceiver = XCVR_INTERNAL;
2315 		cmd->mdio_support = 0;
2316 	}
2317 
2318 	cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2319 	cmd->advertising = from_fw_linkcaps(p->port_type,
2320 					    p->link_cfg.advertising);
2321 	ethtool_cmd_speed_set(cmd,
2322 			      netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2323 	cmd->duplex = DUPLEX_FULL;
2324 	cmd->autoneg = p->link_cfg.autoneg;
2325 	cmd->maxtxpkt = 0;
2326 	cmd->maxrxpkt = 0;
2327 	return 0;
2328 }
2329 
2330 static unsigned int speed_to_caps(int speed)
2331 {
2332 	if (speed == SPEED_100)
2333 		return FW_PORT_CAP_SPEED_100M;
2334 	if (speed == SPEED_1000)
2335 		return FW_PORT_CAP_SPEED_1G;
2336 	if (speed == SPEED_10000)
2337 		return FW_PORT_CAP_SPEED_10G;
2338 	return 0;
2339 }
2340 
2341 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2342 {
2343 	unsigned int cap;
2344 	struct port_info *p = netdev_priv(dev);
2345 	struct link_config *lc = &p->link_cfg;
2346 	u32 speed = ethtool_cmd_speed(cmd);
2347 
2348 	if (cmd->duplex != DUPLEX_FULL)     /* only full-duplex supported */
2349 		return -EINVAL;
2350 
2351 	if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2352 		/*
2353 		 * PHY offers a single speed.  See if that's what's
2354 		 * being requested.
2355 		 */
2356 		if (cmd->autoneg == AUTONEG_DISABLE &&
2357 		    (lc->supported & speed_to_caps(speed)))
2358 			return 0;
2359 		return -EINVAL;
2360 	}
2361 
2362 	if (cmd->autoneg == AUTONEG_DISABLE) {
2363 		cap = speed_to_caps(speed);
2364 
2365 		if (!(lc->supported & cap) || (speed == SPEED_1000) ||
2366 		    (speed == SPEED_10000))
2367 			return -EINVAL;
2368 		lc->requested_speed = cap;
2369 		lc->advertising = 0;
2370 	} else {
2371 		cap = to_fw_linkcaps(cmd->advertising);
2372 		if (!(lc->supported & cap))
2373 			return -EINVAL;
2374 		lc->requested_speed = 0;
2375 		lc->advertising = cap | FW_PORT_CAP_ANEG;
2376 	}
2377 	lc->autoneg = cmd->autoneg;
2378 
2379 	if (netif_running(dev))
2380 		return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2381 				     lc);
2382 	return 0;
2383 }
2384 
2385 static void get_pauseparam(struct net_device *dev,
2386 			   struct ethtool_pauseparam *epause)
2387 {
2388 	struct port_info *p = netdev_priv(dev);
2389 
2390 	epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2391 	epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2392 	epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2393 }
2394 
2395 static int set_pauseparam(struct net_device *dev,
2396 			  struct ethtool_pauseparam *epause)
2397 {
2398 	struct port_info *p = netdev_priv(dev);
2399 	struct link_config *lc = &p->link_cfg;
2400 
2401 	if (epause->autoneg == AUTONEG_DISABLE)
2402 		lc->requested_fc = 0;
2403 	else if (lc->supported & FW_PORT_CAP_ANEG)
2404 		lc->requested_fc = PAUSE_AUTONEG;
2405 	else
2406 		return -EINVAL;
2407 
2408 	if (epause->rx_pause)
2409 		lc->requested_fc |= PAUSE_RX;
2410 	if (epause->tx_pause)
2411 		lc->requested_fc |= PAUSE_TX;
2412 	if (netif_running(dev))
2413 		return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2414 				     lc);
2415 	return 0;
2416 }
2417 
2418 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2419 {
2420 	const struct port_info *pi = netdev_priv(dev);
2421 	const struct sge *s = &pi->adapter->sge;
2422 
2423 	e->rx_max_pending = MAX_RX_BUFFERS;
2424 	e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2425 	e->rx_jumbo_max_pending = 0;
2426 	e->tx_max_pending = MAX_TXQ_ENTRIES;
2427 
2428 	e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2429 	e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2430 	e->rx_jumbo_pending = 0;
2431 	e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2432 }
2433 
2434 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2435 {
2436 	int i;
2437 	const struct port_info *pi = netdev_priv(dev);
2438 	struct adapter *adapter = pi->adapter;
2439 	struct sge *s = &adapter->sge;
2440 
2441 	if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2442 	    e->tx_pending > MAX_TXQ_ENTRIES ||
2443 	    e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2444 	    e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2445 	    e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2446 		return -EINVAL;
2447 
2448 	if (adapter->flags & FULL_INIT_DONE)
2449 		return -EBUSY;
2450 
2451 	for (i = 0; i < pi->nqsets; ++i) {
2452 		s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2453 		s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2454 		s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2455 	}
2456 	return 0;
2457 }
2458 
2459 static int closest_timer(const struct sge *s, int time)
2460 {
2461 	int i, delta, match = 0, min_delta = INT_MAX;
2462 
2463 	for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2464 		delta = time - s->timer_val[i];
2465 		if (delta < 0)
2466 			delta = -delta;
2467 		if (delta < min_delta) {
2468 			min_delta = delta;
2469 			match = i;
2470 		}
2471 	}
2472 	return match;
2473 }
2474 
2475 static int closest_thres(const struct sge *s, int thres)
2476 {
2477 	int i, delta, match = 0, min_delta = INT_MAX;
2478 
2479 	for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2480 		delta = thres - s->counter_val[i];
2481 		if (delta < 0)
2482 			delta = -delta;
2483 		if (delta < min_delta) {
2484 			min_delta = delta;
2485 			match = i;
2486 		}
2487 	}
2488 	return match;
2489 }
2490 
2491 /*
2492  * Return a queue's interrupt hold-off time in us.  0 means no timer.
2493  */
2494 static unsigned int qtimer_val(const struct adapter *adap,
2495 			       const struct sge_rspq *q)
2496 {
2497 	unsigned int idx = q->intr_params >> 1;
2498 
2499 	return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2500 }
2501 
2502 /**
2503  *	set_rxq_intr_params - set a queue's interrupt holdoff parameters
2504  *	@adap: the adapter
2505  *	@q: the Rx queue
2506  *	@us: the hold-off time in us, or 0 to disable timer
2507  *	@cnt: the hold-off packet count, or 0 to disable counter
2508  *
2509  *	Sets an Rx queue's interrupt hold-off time and packet count.  At least
2510  *	one of the two needs to be enabled for the queue to generate interrupts.
2511  */
2512 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
2513 			       unsigned int us, unsigned int cnt)
2514 {
2515 	if ((us | cnt) == 0)
2516 		cnt = 1;
2517 
2518 	if (cnt) {
2519 		int err;
2520 		u32 v, new_idx;
2521 
2522 		new_idx = closest_thres(&adap->sge, cnt);
2523 		if (q->desc && q->pktcnt_idx != new_idx) {
2524 			/* the queue has already been created, update it */
2525 			v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2526 			    FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2527 			    FW_PARAMS_PARAM_YZ(q->cntxt_id);
2528 			err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2529 					    &new_idx);
2530 			if (err)
2531 				return err;
2532 		}
2533 		q->pktcnt_idx = new_idx;
2534 	}
2535 
2536 	us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2537 	q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2538 	return 0;
2539 }
2540 
2541 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2542 {
2543 	const struct port_info *pi = netdev_priv(dev);
2544 	struct adapter *adap = pi->adapter;
2545 	struct sge_rspq *q;
2546 	int i;
2547 	int r = 0;
2548 
2549 	for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2550 		q = &adap->sge.ethrxq[i].rspq;
2551 		r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2552 			c->rx_max_coalesced_frames);
2553 		if (r) {
2554 			dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2555 			break;
2556 		}
2557 	}
2558 	return r;
2559 }
2560 
2561 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2562 {
2563 	const struct port_info *pi = netdev_priv(dev);
2564 	const struct adapter *adap = pi->adapter;
2565 	const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2566 
2567 	c->rx_coalesce_usecs = qtimer_val(adap, rq);
2568 	c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2569 		adap->sge.counter_val[rq->pktcnt_idx] : 0;
2570 	return 0;
2571 }
2572 
2573 /**
2574  *	eeprom_ptov - translate a physical EEPROM address to virtual
2575  *	@phys_addr: the physical EEPROM address
2576  *	@fn: the PCI function number
2577  *	@sz: size of function-specific area
2578  *
2579  *	Translate a physical EEPROM address to virtual.  The first 1K is
2580  *	accessed through virtual addresses starting at 31K, the rest is
2581  *	accessed through virtual addresses starting at 0.
2582  *
2583  *	The mapping is as follows:
2584  *	[0..1K) -> [31K..32K)
2585  *	[1K..1K+A) -> [31K-A..31K)
2586  *	[1K+A..ES) -> [0..ES-A-1K)
2587  *
2588  *	where A = @fn * @sz, and ES = EEPROM size.
2589  */
2590 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2591 {
2592 	fn *= sz;
2593 	if (phys_addr < 1024)
2594 		return phys_addr + (31 << 10);
2595 	if (phys_addr < 1024 + fn)
2596 		return 31744 - fn + phys_addr - 1024;
2597 	if (phys_addr < EEPROMSIZE)
2598 		return phys_addr - 1024 - fn;
2599 	return -EINVAL;
2600 }
2601 
2602 /*
2603  * The next two routines implement eeprom read/write from physical addresses.
2604  */
2605 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2606 {
2607 	int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2608 
2609 	if (vaddr >= 0)
2610 		vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2611 	return vaddr < 0 ? vaddr : 0;
2612 }
2613 
2614 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2615 {
2616 	int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2617 
2618 	if (vaddr >= 0)
2619 		vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2620 	return vaddr < 0 ? vaddr : 0;
2621 }
2622 
2623 #define EEPROM_MAGIC 0x38E2F10C
2624 
2625 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2626 		      u8 *data)
2627 {
2628 	int i, err = 0;
2629 	struct adapter *adapter = netdev2adap(dev);
2630 
2631 	u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2632 	if (!buf)
2633 		return -ENOMEM;
2634 
2635 	e->magic = EEPROM_MAGIC;
2636 	for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2637 		err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2638 
2639 	if (!err)
2640 		memcpy(data, buf + e->offset, e->len);
2641 	kfree(buf);
2642 	return err;
2643 }
2644 
2645 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2646 		      u8 *data)
2647 {
2648 	u8 *buf;
2649 	int err = 0;
2650 	u32 aligned_offset, aligned_len, *p;
2651 	struct adapter *adapter = netdev2adap(dev);
2652 
2653 	if (eeprom->magic != EEPROM_MAGIC)
2654 		return -EINVAL;
2655 
2656 	aligned_offset = eeprom->offset & ~3;
2657 	aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2658 
2659 	if (adapter->fn > 0) {
2660 		u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2661 
2662 		if (aligned_offset < start ||
2663 		    aligned_offset + aligned_len > start + EEPROMPFSIZE)
2664 			return -EPERM;
2665 	}
2666 
2667 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2668 		/*
2669 		 * RMW possibly needed for first or last words.
2670 		 */
2671 		buf = kmalloc(aligned_len, GFP_KERNEL);
2672 		if (!buf)
2673 			return -ENOMEM;
2674 		err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2675 		if (!err && aligned_len > 4)
2676 			err = eeprom_rd_phys(adapter,
2677 					     aligned_offset + aligned_len - 4,
2678 					     (u32 *)&buf[aligned_len - 4]);
2679 		if (err)
2680 			goto out;
2681 		memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2682 	} else
2683 		buf = data;
2684 
2685 	err = t4_seeprom_wp(adapter, false);
2686 	if (err)
2687 		goto out;
2688 
2689 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2690 		err = eeprom_wr_phys(adapter, aligned_offset, *p);
2691 		aligned_offset += 4;
2692 	}
2693 
2694 	if (!err)
2695 		err = t4_seeprom_wp(adapter, true);
2696 out:
2697 	if (buf != data)
2698 		kfree(buf);
2699 	return err;
2700 }
2701 
2702 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2703 {
2704 	int ret;
2705 	const struct firmware *fw;
2706 	struct adapter *adap = netdev2adap(netdev);
2707 
2708 	ef->data[sizeof(ef->data) - 1] = '\0';
2709 	ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2710 	if (ret < 0)
2711 		return ret;
2712 
2713 	ret = t4_load_fw(adap, fw->data, fw->size);
2714 	release_firmware(fw);
2715 	if (!ret)
2716 		dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2717 	return ret;
2718 }
2719 
2720 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2721 #define BCAST_CRC 0xa0ccc1a6
2722 
2723 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2724 {
2725 	wol->supported = WAKE_BCAST | WAKE_MAGIC;
2726 	wol->wolopts = netdev2adap(dev)->wol;
2727 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2728 }
2729 
2730 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2731 {
2732 	int err = 0;
2733 	struct port_info *pi = netdev_priv(dev);
2734 
2735 	if (wol->wolopts & ~WOL_SUPPORTED)
2736 		return -EINVAL;
2737 	t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2738 			    (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2739 	if (wol->wolopts & WAKE_BCAST) {
2740 		err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2741 					~0ULL, 0, false);
2742 		if (!err)
2743 			err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2744 						~6ULL, ~0ULL, BCAST_CRC, true);
2745 	} else
2746 		t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2747 	return err;
2748 }
2749 
2750 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2751 {
2752 	const struct port_info *pi = netdev_priv(dev);
2753 	netdev_features_t changed = dev->features ^ features;
2754 	int err;
2755 
2756 	if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
2757 		return 0;
2758 
2759 	err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2760 			    -1, -1, -1,
2761 			    !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
2762 	if (unlikely(err))
2763 		dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
2764 	return err;
2765 }
2766 
2767 static u32 get_rss_table_size(struct net_device *dev)
2768 {
2769 	const struct port_info *pi = netdev_priv(dev);
2770 
2771 	return pi->rss_size;
2772 }
2773 
2774 static int get_rss_table(struct net_device *dev, u32 *p)
2775 {
2776 	const struct port_info *pi = netdev_priv(dev);
2777 	unsigned int n = pi->rss_size;
2778 
2779 	while (n--)
2780 		p[n] = pi->rss[n];
2781 	return 0;
2782 }
2783 
2784 static int set_rss_table(struct net_device *dev, const u32 *p)
2785 {
2786 	unsigned int i;
2787 	struct port_info *pi = netdev_priv(dev);
2788 
2789 	for (i = 0; i < pi->rss_size; i++)
2790 		pi->rss[i] = p[i];
2791 	if (pi->adapter->flags & FULL_INIT_DONE)
2792 		return write_rss(pi, pi->rss);
2793 	return 0;
2794 }
2795 
2796 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2797 		     u32 *rules)
2798 {
2799 	const struct port_info *pi = netdev_priv(dev);
2800 
2801 	switch (info->cmd) {
2802 	case ETHTOOL_GRXFH: {
2803 		unsigned int v = pi->rss_mode;
2804 
2805 		info->data = 0;
2806 		switch (info->flow_type) {
2807 		case TCP_V4_FLOW:
2808 			if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2809 				info->data = RXH_IP_SRC | RXH_IP_DST |
2810 					     RXH_L4_B_0_1 | RXH_L4_B_2_3;
2811 			else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2812 				info->data = RXH_IP_SRC | RXH_IP_DST;
2813 			break;
2814 		case UDP_V4_FLOW:
2815 			if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2816 			    (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2817 				info->data = RXH_IP_SRC | RXH_IP_DST |
2818 					     RXH_L4_B_0_1 | RXH_L4_B_2_3;
2819 			else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2820 				info->data = RXH_IP_SRC | RXH_IP_DST;
2821 			break;
2822 		case SCTP_V4_FLOW:
2823 		case AH_ESP_V4_FLOW:
2824 		case IPV4_FLOW:
2825 			if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2826 				info->data = RXH_IP_SRC | RXH_IP_DST;
2827 			break;
2828 		case TCP_V6_FLOW:
2829 			if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2830 				info->data = RXH_IP_SRC | RXH_IP_DST |
2831 					     RXH_L4_B_0_1 | RXH_L4_B_2_3;
2832 			else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2833 				info->data = RXH_IP_SRC | RXH_IP_DST;
2834 			break;
2835 		case UDP_V6_FLOW:
2836 			if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2837 			    (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2838 				info->data = RXH_IP_SRC | RXH_IP_DST |
2839 					     RXH_L4_B_0_1 | RXH_L4_B_2_3;
2840 			else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2841 				info->data = RXH_IP_SRC | RXH_IP_DST;
2842 			break;
2843 		case SCTP_V6_FLOW:
2844 		case AH_ESP_V6_FLOW:
2845 		case IPV6_FLOW:
2846 			if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2847 				info->data = RXH_IP_SRC | RXH_IP_DST;
2848 			break;
2849 		}
2850 		return 0;
2851 	}
2852 	case ETHTOOL_GRXRINGS:
2853 		info->data = pi->nqsets;
2854 		return 0;
2855 	}
2856 	return -EOPNOTSUPP;
2857 }
2858 
2859 static const struct ethtool_ops cxgb_ethtool_ops = {
2860 	.get_settings      = get_settings,
2861 	.set_settings      = set_settings,
2862 	.get_drvinfo       = get_drvinfo,
2863 	.get_msglevel      = get_msglevel,
2864 	.set_msglevel      = set_msglevel,
2865 	.get_ringparam     = get_sge_param,
2866 	.set_ringparam     = set_sge_param,
2867 	.get_coalesce      = get_coalesce,
2868 	.set_coalesce      = set_coalesce,
2869 	.get_eeprom_len    = get_eeprom_len,
2870 	.get_eeprom        = get_eeprom,
2871 	.set_eeprom        = set_eeprom,
2872 	.get_pauseparam    = get_pauseparam,
2873 	.set_pauseparam    = set_pauseparam,
2874 	.get_link          = ethtool_op_get_link,
2875 	.get_strings       = get_strings,
2876 	.set_phys_id       = identify_port,
2877 	.nway_reset        = restart_autoneg,
2878 	.get_sset_count    = get_sset_count,
2879 	.get_ethtool_stats = get_stats,
2880 	.get_regs_len      = get_regs_len,
2881 	.get_regs          = get_regs,
2882 	.get_wol           = get_wol,
2883 	.set_wol           = set_wol,
2884 	.get_rxnfc         = get_rxnfc,
2885 	.get_rxfh_indir_size = get_rss_table_size,
2886 	.get_rxfh_indir    = get_rss_table,
2887 	.set_rxfh_indir    = set_rss_table,
2888 	.flash_device      = set_flash,
2889 };
2890 
2891 /*
2892  * debugfs support
2893  */
2894 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2895 			loff_t *ppos)
2896 {
2897 	loff_t pos = *ppos;
2898 	loff_t avail = file_inode(file)->i_size;
2899 	unsigned int mem = (uintptr_t)file->private_data & 3;
2900 	struct adapter *adap = file->private_data - mem;
2901 
2902 	if (pos < 0)
2903 		return -EINVAL;
2904 	if (pos >= avail)
2905 		return 0;
2906 	if (count > avail - pos)
2907 		count = avail - pos;
2908 
2909 	while (count) {
2910 		size_t len;
2911 		int ret, ofst;
2912 		__be32 data[16];
2913 
2914 		if ((mem == MEM_MC) || (mem == MEM_MC1))
2915 			ret = t4_mc_read(adap, mem % MEM_MC, pos, data, NULL);
2916 		else
2917 			ret = t4_edc_read(adap, mem, pos, data, NULL);
2918 		if (ret)
2919 			return ret;
2920 
2921 		ofst = pos % sizeof(data);
2922 		len = min(count, sizeof(data) - ofst);
2923 		if (copy_to_user(buf, (u8 *)data + ofst, len))
2924 			return -EFAULT;
2925 
2926 		buf += len;
2927 		pos += len;
2928 		count -= len;
2929 	}
2930 	count = pos - *ppos;
2931 	*ppos = pos;
2932 	return count;
2933 }
2934 
2935 static const struct file_operations mem_debugfs_fops = {
2936 	.owner   = THIS_MODULE,
2937 	.open    = simple_open,
2938 	.read    = mem_read,
2939 	.llseek  = default_llseek,
2940 };
2941 
2942 static void add_debugfs_mem(struct adapter *adap, const char *name,
2943 			    unsigned int idx, unsigned int size_mb)
2944 {
2945 	struct dentry *de;
2946 
2947 	de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2948 				 (void *)adap + idx, &mem_debugfs_fops);
2949 	if (de && de->d_inode)
2950 		de->d_inode->i_size = size_mb << 20;
2951 }
2952 
2953 static int setup_debugfs(struct adapter *adap)
2954 {
2955 	int i;
2956 	u32 size;
2957 
2958 	if (IS_ERR_OR_NULL(adap->debugfs_root))
2959 		return -1;
2960 
2961 	i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2962 	if (i & EDRAM0_ENABLE) {
2963 		size = t4_read_reg(adap, MA_EDRAM0_BAR);
2964 		add_debugfs_mem(adap, "edc0", MEM_EDC0,	EDRAM_SIZE_GET(size));
2965 	}
2966 	if (i & EDRAM1_ENABLE) {
2967 		size = t4_read_reg(adap, MA_EDRAM1_BAR);
2968 		add_debugfs_mem(adap, "edc1", MEM_EDC1, EDRAM_SIZE_GET(size));
2969 	}
2970 	if (is_t4(adap->chip)) {
2971 		size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2972 		if (i & EXT_MEM_ENABLE)
2973 			add_debugfs_mem(adap, "mc", MEM_MC,
2974 					EXT_MEM_SIZE_GET(size));
2975 	} else {
2976 		if (i & EXT_MEM_ENABLE) {
2977 			size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2978 			add_debugfs_mem(adap, "mc0", MEM_MC0,
2979 					EXT_MEM_SIZE_GET(size));
2980 		}
2981 		if (i & EXT_MEM1_ENABLE) {
2982 			size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR);
2983 			add_debugfs_mem(adap, "mc1", MEM_MC1,
2984 					EXT_MEM_SIZE_GET(size));
2985 		}
2986 	}
2987 	if (adap->l2t)
2988 		debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2989 				    &t4_l2t_fops);
2990 	return 0;
2991 }
2992 
2993 /*
2994  * upper-layer driver support
2995  */
2996 
2997 /*
2998  * Allocate an active-open TID and set it to the supplied value.
2999  */
3000 int cxgb4_alloc_atid(struct tid_info *t, void *data)
3001 {
3002 	int atid = -1;
3003 
3004 	spin_lock_bh(&t->atid_lock);
3005 	if (t->afree) {
3006 		union aopen_entry *p = t->afree;
3007 
3008 		atid = (p - t->atid_tab) + t->atid_base;
3009 		t->afree = p->next;
3010 		p->data = data;
3011 		t->atids_in_use++;
3012 	}
3013 	spin_unlock_bh(&t->atid_lock);
3014 	return atid;
3015 }
3016 EXPORT_SYMBOL(cxgb4_alloc_atid);
3017 
3018 /*
3019  * Release an active-open TID.
3020  */
3021 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
3022 {
3023 	union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
3024 
3025 	spin_lock_bh(&t->atid_lock);
3026 	p->next = t->afree;
3027 	t->afree = p;
3028 	t->atids_in_use--;
3029 	spin_unlock_bh(&t->atid_lock);
3030 }
3031 EXPORT_SYMBOL(cxgb4_free_atid);
3032 
3033 /*
3034  * Allocate a server TID and set it to the supplied value.
3035  */
3036 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
3037 {
3038 	int stid;
3039 
3040 	spin_lock_bh(&t->stid_lock);
3041 	if (family == PF_INET) {
3042 		stid = find_first_zero_bit(t->stid_bmap, t->nstids);
3043 		if (stid < t->nstids)
3044 			__set_bit(stid, t->stid_bmap);
3045 		else
3046 			stid = -1;
3047 	} else {
3048 		stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
3049 		if (stid < 0)
3050 			stid = -1;
3051 	}
3052 	if (stid >= 0) {
3053 		t->stid_tab[stid].data = data;
3054 		stid += t->stid_base;
3055 		t->stids_in_use++;
3056 	}
3057 	spin_unlock_bh(&t->stid_lock);
3058 	return stid;
3059 }
3060 EXPORT_SYMBOL(cxgb4_alloc_stid);
3061 
3062 /* Allocate a server filter TID and set it to the supplied value.
3063  */
3064 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
3065 {
3066 	int stid;
3067 
3068 	spin_lock_bh(&t->stid_lock);
3069 	if (family == PF_INET) {
3070 		stid = find_next_zero_bit(t->stid_bmap,
3071 				t->nstids + t->nsftids, t->nstids);
3072 		if (stid < (t->nstids + t->nsftids))
3073 			__set_bit(stid, t->stid_bmap);
3074 		else
3075 			stid = -1;
3076 	} else {
3077 		stid = -1;
3078 	}
3079 	if (stid >= 0) {
3080 		t->stid_tab[stid].data = data;
3081 		stid += t->stid_base;
3082 		t->stids_in_use++;
3083 	}
3084 	spin_unlock_bh(&t->stid_lock);
3085 	return stid;
3086 }
3087 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3088 
3089 /* Release a server TID.
3090  */
3091 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3092 {
3093 	stid -= t->stid_base;
3094 	spin_lock_bh(&t->stid_lock);
3095 	if (family == PF_INET)
3096 		__clear_bit(stid, t->stid_bmap);
3097 	else
3098 		bitmap_release_region(t->stid_bmap, stid, 2);
3099 	t->stid_tab[stid].data = NULL;
3100 	t->stids_in_use--;
3101 	spin_unlock_bh(&t->stid_lock);
3102 }
3103 EXPORT_SYMBOL(cxgb4_free_stid);
3104 
3105 /*
3106  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
3107  */
3108 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3109 			   unsigned int tid)
3110 {
3111 	struct cpl_tid_release *req;
3112 
3113 	set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3114 	req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3115 	INIT_TP_WR(req, tid);
3116 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3117 }
3118 
3119 /*
3120  * Queue a TID release request and if necessary schedule a work queue to
3121  * process it.
3122  */
3123 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3124 				    unsigned int tid)
3125 {
3126 	void **p = &t->tid_tab[tid];
3127 	struct adapter *adap = container_of(t, struct adapter, tids);
3128 
3129 	spin_lock_bh(&adap->tid_release_lock);
3130 	*p = adap->tid_release_head;
3131 	/* Low 2 bits encode the Tx channel number */
3132 	adap->tid_release_head = (void **)((uintptr_t)p | chan);
3133 	if (!adap->tid_release_task_busy) {
3134 		adap->tid_release_task_busy = true;
3135 		queue_work(workq, &adap->tid_release_task);
3136 	}
3137 	spin_unlock_bh(&adap->tid_release_lock);
3138 }
3139 
3140 /*
3141  * Process the list of pending TID release requests.
3142  */
3143 static void process_tid_release_list(struct work_struct *work)
3144 {
3145 	struct sk_buff *skb;
3146 	struct adapter *adap;
3147 
3148 	adap = container_of(work, struct adapter, tid_release_task);
3149 
3150 	spin_lock_bh(&adap->tid_release_lock);
3151 	while (adap->tid_release_head) {
3152 		void **p = adap->tid_release_head;
3153 		unsigned int chan = (uintptr_t)p & 3;
3154 		p = (void *)p - chan;
3155 
3156 		adap->tid_release_head = *p;
3157 		*p = NULL;
3158 		spin_unlock_bh(&adap->tid_release_lock);
3159 
3160 		while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3161 					 GFP_KERNEL)))
3162 			schedule_timeout_uninterruptible(1);
3163 
3164 		mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3165 		t4_ofld_send(adap, skb);
3166 		spin_lock_bh(&adap->tid_release_lock);
3167 	}
3168 	adap->tid_release_task_busy = false;
3169 	spin_unlock_bh(&adap->tid_release_lock);
3170 }
3171 
3172 /*
3173  * Release a TID and inform HW.  If we are unable to allocate the release
3174  * message we defer to a work queue.
3175  */
3176 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3177 {
3178 	void *old;
3179 	struct sk_buff *skb;
3180 	struct adapter *adap = container_of(t, struct adapter, tids);
3181 
3182 	old = t->tid_tab[tid];
3183 	skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3184 	if (likely(skb)) {
3185 		t->tid_tab[tid] = NULL;
3186 		mk_tid_release(skb, chan, tid);
3187 		t4_ofld_send(adap, skb);
3188 	} else
3189 		cxgb4_queue_tid_release(t, chan, tid);
3190 	if (old)
3191 		atomic_dec(&t->tids_in_use);
3192 }
3193 EXPORT_SYMBOL(cxgb4_remove_tid);
3194 
3195 /*
3196  * Allocate and initialize the TID tables.  Returns 0 on success.
3197  */
3198 static int tid_init(struct tid_info *t)
3199 {
3200 	size_t size;
3201 	unsigned int stid_bmap_size;
3202 	unsigned int natids = t->natids;
3203 
3204 	stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3205 	size = t->ntids * sizeof(*t->tid_tab) +
3206 	       natids * sizeof(*t->atid_tab) +
3207 	       t->nstids * sizeof(*t->stid_tab) +
3208 	       t->nsftids * sizeof(*t->stid_tab) +
3209 	       stid_bmap_size * sizeof(long) +
3210 	       t->nftids * sizeof(*t->ftid_tab) +
3211 	       t->nsftids * sizeof(*t->ftid_tab);
3212 
3213 	t->tid_tab = t4_alloc_mem(size);
3214 	if (!t->tid_tab)
3215 		return -ENOMEM;
3216 
3217 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3218 	t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3219 	t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3220 	t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3221 	spin_lock_init(&t->stid_lock);
3222 	spin_lock_init(&t->atid_lock);
3223 
3224 	t->stids_in_use = 0;
3225 	t->afree = NULL;
3226 	t->atids_in_use = 0;
3227 	atomic_set(&t->tids_in_use, 0);
3228 
3229 	/* Setup the free list for atid_tab and clear the stid bitmap. */
3230 	if (natids) {
3231 		while (--natids)
3232 			t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3233 		t->afree = t->atid_tab;
3234 	}
3235 	bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3236 	return 0;
3237 }
3238 
3239 static int cxgb4_clip_get(const struct net_device *dev,
3240 			  const struct in6_addr *lip)
3241 {
3242 	struct adapter *adap;
3243 	struct fw_clip_cmd c;
3244 
3245 	adap = netdev2adap(dev);
3246 	memset(&c, 0, sizeof(c));
3247 	c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3248 			FW_CMD_REQUEST | FW_CMD_WRITE);
3249 	c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
3250 	*(__be64 *)&c.ip_hi = *(__be64 *)(lip->s6_addr);
3251 	*(__be64 *)&c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3252 	return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3253 }
3254 
3255 static int cxgb4_clip_release(const struct net_device *dev,
3256 			      const struct in6_addr *lip)
3257 {
3258 	struct adapter *adap;
3259 	struct fw_clip_cmd c;
3260 
3261 	adap = netdev2adap(dev);
3262 	memset(&c, 0, sizeof(c));
3263 	c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3264 			FW_CMD_REQUEST | FW_CMD_READ);
3265 	c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
3266 	*(__be64 *)&c.ip_hi = *(__be64 *)(lip->s6_addr);
3267 	*(__be64 *)&c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3268 	return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3269 }
3270 
3271 /**
3272  *	cxgb4_create_server - create an IP server
3273  *	@dev: the device
3274  *	@stid: the server TID
3275  *	@sip: local IP address to bind server to
3276  *	@sport: the server's TCP port
3277  *	@queue: queue to direct messages from this server to
3278  *
3279  *	Create an IP server for the given port and address.
3280  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
3281  */
3282 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3283 			__be32 sip, __be16 sport, __be16 vlan,
3284 			unsigned int queue)
3285 {
3286 	unsigned int chan;
3287 	struct sk_buff *skb;
3288 	struct adapter *adap;
3289 	struct cpl_pass_open_req *req;
3290 	int ret;
3291 
3292 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3293 	if (!skb)
3294 		return -ENOMEM;
3295 
3296 	adap = netdev2adap(dev);
3297 	req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3298 	INIT_TP_WR(req, 0);
3299 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3300 	req->local_port = sport;
3301 	req->peer_port = htons(0);
3302 	req->local_ip = sip;
3303 	req->peer_ip = htonl(0);
3304 	chan = rxq_to_chan(&adap->sge, queue);
3305 	req->opt0 = cpu_to_be64(TX_CHAN(chan));
3306 	req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3307 				SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3308 	ret = t4_mgmt_tx(adap, skb);
3309 	return net_xmit_eval(ret);
3310 }
3311 EXPORT_SYMBOL(cxgb4_create_server);
3312 
3313 /*	cxgb4_create_server6 - create an IPv6 server
3314  *	@dev: the device
3315  *	@stid: the server TID
3316  *	@sip: local IPv6 address to bind server to
3317  *	@sport: the server's TCP port
3318  *	@queue: queue to direct messages from this server to
3319  *
3320  *	Create an IPv6 server for the given port and address.
3321  *	Returns <0 on error and one of the %NET_XMIT_* values on success.
3322  */
3323 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
3324 			 const struct in6_addr *sip, __be16 sport,
3325 			 unsigned int queue)
3326 {
3327 	unsigned int chan;
3328 	struct sk_buff *skb;
3329 	struct adapter *adap;
3330 	struct cpl_pass_open_req6 *req;
3331 	int ret;
3332 
3333 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3334 	if (!skb)
3335 		return -ENOMEM;
3336 
3337 	adap = netdev2adap(dev);
3338 	req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
3339 	INIT_TP_WR(req, 0);
3340 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
3341 	req->local_port = sport;
3342 	req->peer_port = htons(0);
3343 	req->local_ip_hi = *(__be64 *)(sip->s6_addr);
3344 	req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
3345 	req->peer_ip_hi = cpu_to_be64(0);
3346 	req->peer_ip_lo = cpu_to_be64(0);
3347 	chan = rxq_to_chan(&adap->sge, queue);
3348 	req->opt0 = cpu_to_be64(TX_CHAN(chan));
3349 	req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3350 				SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3351 	ret = t4_mgmt_tx(adap, skb);
3352 	return net_xmit_eval(ret);
3353 }
3354 EXPORT_SYMBOL(cxgb4_create_server6);
3355 
3356 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
3357 			unsigned int queue, bool ipv6)
3358 {
3359 	struct sk_buff *skb;
3360 	struct adapter *adap;
3361 	struct cpl_close_listsvr_req *req;
3362 	int ret;
3363 
3364 	adap = netdev2adap(dev);
3365 
3366 	skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3367 	if (!skb)
3368 		return -ENOMEM;
3369 
3370 	req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
3371 	INIT_TP_WR(req, 0);
3372 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
3373 	req->reply_ctrl = htons(NO_REPLY(0) | (ipv6 ? LISTSVR_IPV6(1) :
3374 				LISTSVR_IPV6(0)) | QUEUENO(queue));
3375 	ret = t4_mgmt_tx(adap, skb);
3376 	return net_xmit_eval(ret);
3377 }
3378 EXPORT_SYMBOL(cxgb4_remove_server);
3379 
3380 /**
3381  *	cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3382  *	@mtus: the HW MTU table
3383  *	@mtu: the target MTU
3384  *	@idx: index of selected entry in the MTU table
3385  *
3386  *	Returns the index and the value in the HW MTU table that is closest to
3387  *	but does not exceed @mtu, unless @mtu is smaller than any value in the
3388  *	table, in which case that smallest available value is selected.
3389  */
3390 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3391 			    unsigned int *idx)
3392 {
3393 	unsigned int i = 0;
3394 
3395 	while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3396 		++i;
3397 	if (idx)
3398 		*idx = i;
3399 	return mtus[i];
3400 }
3401 EXPORT_SYMBOL(cxgb4_best_mtu);
3402 
3403 /**
3404  *	cxgb4_port_chan - get the HW channel of a port
3405  *	@dev: the net device for the port
3406  *
3407  *	Return the HW Tx channel of the given port.
3408  */
3409 unsigned int cxgb4_port_chan(const struct net_device *dev)
3410 {
3411 	return netdev2pinfo(dev)->tx_chan;
3412 }
3413 EXPORT_SYMBOL(cxgb4_port_chan);
3414 
3415 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3416 {
3417 	struct adapter *adap = netdev2adap(dev);
3418 	u32 v1, v2, lp_count, hp_count;
3419 
3420 	v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3421 	v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3422 	if (is_t4(adap->chip)) {
3423 		lp_count = G_LP_COUNT(v1);
3424 		hp_count = G_HP_COUNT(v1);
3425 	} else {
3426 		lp_count = G_LP_COUNT_T5(v1);
3427 		hp_count = G_HP_COUNT_T5(v2);
3428 	}
3429 	return lpfifo ? lp_count : hp_count;
3430 }
3431 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3432 
3433 /**
3434  *	cxgb4_port_viid - get the VI id of a port
3435  *	@dev: the net device for the port
3436  *
3437  *	Return the VI id of the given port.
3438  */
3439 unsigned int cxgb4_port_viid(const struct net_device *dev)
3440 {
3441 	return netdev2pinfo(dev)->viid;
3442 }
3443 EXPORT_SYMBOL(cxgb4_port_viid);
3444 
3445 /**
3446  *	cxgb4_port_idx - get the index of a port
3447  *	@dev: the net device for the port
3448  *
3449  *	Return the index of the given port.
3450  */
3451 unsigned int cxgb4_port_idx(const struct net_device *dev)
3452 {
3453 	return netdev2pinfo(dev)->port_id;
3454 }
3455 EXPORT_SYMBOL(cxgb4_port_idx);
3456 
3457 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3458 			 struct tp_tcp_stats *v6)
3459 {
3460 	struct adapter *adap = pci_get_drvdata(pdev);
3461 
3462 	spin_lock(&adap->stats_lock);
3463 	t4_tp_get_tcp_stats(adap, v4, v6);
3464 	spin_unlock(&adap->stats_lock);
3465 }
3466 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3467 
3468 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3469 		      const unsigned int *pgsz_order)
3470 {
3471 	struct adapter *adap = netdev2adap(dev);
3472 
3473 	t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3474 	t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3475 		     HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3476 		     HPZ3(pgsz_order[3]));
3477 }
3478 EXPORT_SYMBOL(cxgb4_iscsi_init);
3479 
3480 int cxgb4_flush_eq_cache(struct net_device *dev)
3481 {
3482 	struct adapter *adap = netdev2adap(dev);
3483 	int ret;
3484 
3485 	ret = t4_fwaddrspace_write(adap, adap->mbox,
3486 				   0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3487 	return ret;
3488 }
3489 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3490 
3491 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3492 {
3493 	u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3494 	__be64 indices;
3495 	int ret;
3496 
3497 	ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3498 	if (!ret) {
3499 		*cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3500 		*pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3501 	}
3502 	return ret;
3503 }
3504 
3505 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3506 			u16 size)
3507 {
3508 	struct adapter *adap = netdev2adap(dev);
3509 	u16 hw_pidx, hw_cidx;
3510 	int ret;
3511 
3512 	ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3513 	if (ret)
3514 		goto out;
3515 
3516 	if (pidx != hw_pidx) {
3517 		u16 delta;
3518 
3519 		if (pidx >= hw_pidx)
3520 			delta = pidx - hw_pidx;
3521 		else
3522 			delta = size - hw_pidx + pidx;
3523 		wmb();
3524 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3525 			     QID(qid) | PIDX(delta));
3526 	}
3527 out:
3528 	return ret;
3529 }
3530 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3531 
3532 void cxgb4_disable_db_coalescing(struct net_device *dev)
3533 {
3534 	struct adapter *adap;
3535 
3536 	adap = netdev2adap(dev);
3537 	t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE,
3538 			 F_NOCOALESCE);
3539 }
3540 EXPORT_SYMBOL(cxgb4_disable_db_coalescing);
3541 
3542 void cxgb4_enable_db_coalescing(struct net_device *dev)
3543 {
3544 	struct adapter *adap;
3545 
3546 	adap = netdev2adap(dev);
3547 	t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE, 0);
3548 }
3549 EXPORT_SYMBOL(cxgb4_enable_db_coalescing);
3550 
3551 static struct pci_driver cxgb4_driver;
3552 
3553 static void check_neigh_update(struct neighbour *neigh)
3554 {
3555 	const struct device *parent;
3556 	const struct net_device *netdev = neigh->dev;
3557 
3558 	if (netdev->priv_flags & IFF_802_1Q_VLAN)
3559 		netdev = vlan_dev_real_dev(netdev);
3560 	parent = netdev->dev.parent;
3561 	if (parent && parent->driver == &cxgb4_driver.driver)
3562 		t4_l2t_update(dev_get_drvdata(parent), neigh);
3563 }
3564 
3565 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3566 		       void *data)
3567 {
3568 	switch (event) {
3569 	case NETEVENT_NEIGH_UPDATE:
3570 		check_neigh_update(data);
3571 		break;
3572 	case NETEVENT_REDIRECT:
3573 	default:
3574 		break;
3575 	}
3576 	return 0;
3577 }
3578 
3579 static bool netevent_registered;
3580 static struct notifier_block cxgb4_netevent_nb = {
3581 	.notifier_call = netevent_cb
3582 };
3583 
3584 static void drain_db_fifo(struct adapter *adap, int usecs)
3585 {
3586 	u32 v1, v2, lp_count, hp_count;
3587 
3588 	do {
3589 		v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3590 		v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3591 		if (is_t4(adap->chip)) {
3592 			lp_count = G_LP_COUNT(v1);
3593 			hp_count = G_HP_COUNT(v1);
3594 		} else {
3595 			lp_count = G_LP_COUNT_T5(v1);
3596 			hp_count = G_HP_COUNT_T5(v2);
3597 		}
3598 
3599 		if (lp_count == 0 && hp_count == 0)
3600 			break;
3601 		set_current_state(TASK_UNINTERRUPTIBLE);
3602 		schedule_timeout(usecs_to_jiffies(usecs));
3603 	} while (1);
3604 }
3605 
3606 static void disable_txq_db(struct sge_txq *q)
3607 {
3608 	spin_lock_irq(&q->db_lock);
3609 	q->db_disabled = 1;
3610 	spin_unlock_irq(&q->db_lock);
3611 }
3612 
3613 static void enable_txq_db(struct sge_txq *q)
3614 {
3615 	spin_lock_irq(&q->db_lock);
3616 	q->db_disabled = 0;
3617 	spin_unlock_irq(&q->db_lock);
3618 }
3619 
3620 static void disable_dbs(struct adapter *adap)
3621 {
3622 	int i;
3623 
3624 	for_each_ethrxq(&adap->sge, i)
3625 		disable_txq_db(&adap->sge.ethtxq[i].q);
3626 	for_each_ofldrxq(&adap->sge, i)
3627 		disable_txq_db(&adap->sge.ofldtxq[i].q);
3628 	for_each_port(adap, i)
3629 		disable_txq_db(&adap->sge.ctrlq[i].q);
3630 }
3631 
3632 static void enable_dbs(struct adapter *adap)
3633 {
3634 	int i;
3635 
3636 	for_each_ethrxq(&adap->sge, i)
3637 		enable_txq_db(&adap->sge.ethtxq[i].q);
3638 	for_each_ofldrxq(&adap->sge, i)
3639 		enable_txq_db(&adap->sge.ofldtxq[i].q);
3640 	for_each_port(adap, i)
3641 		enable_txq_db(&adap->sge.ctrlq[i].q);
3642 }
3643 
3644 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3645 {
3646 	u16 hw_pidx, hw_cidx;
3647 	int ret;
3648 
3649 	spin_lock_bh(&q->db_lock);
3650 	ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3651 	if (ret)
3652 		goto out;
3653 	if (q->db_pidx != hw_pidx) {
3654 		u16 delta;
3655 
3656 		if (q->db_pidx >= hw_pidx)
3657 			delta = q->db_pidx - hw_pidx;
3658 		else
3659 			delta = q->size - hw_pidx + q->db_pidx;
3660 		wmb();
3661 		t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3662 			     QID(q->cntxt_id) | PIDX(delta));
3663 	}
3664 out:
3665 	q->db_disabled = 0;
3666 	spin_unlock_bh(&q->db_lock);
3667 	if (ret)
3668 		CH_WARN(adap, "DB drop recovery failed.\n");
3669 }
3670 static void recover_all_queues(struct adapter *adap)
3671 {
3672 	int i;
3673 
3674 	for_each_ethrxq(&adap->sge, i)
3675 		sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3676 	for_each_ofldrxq(&adap->sge, i)
3677 		sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3678 	for_each_port(adap, i)
3679 		sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3680 }
3681 
3682 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3683 {
3684 	mutex_lock(&uld_mutex);
3685 	if (adap->uld_handle[CXGB4_ULD_RDMA])
3686 		ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3687 				cmd);
3688 	mutex_unlock(&uld_mutex);
3689 }
3690 
3691 static void process_db_full(struct work_struct *work)
3692 {
3693 	struct adapter *adap;
3694 
3695 	adap = container_of(work, struct adapter, db_full_task);
3696 
3697 	notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3698 	drain_db_fifo(adap, dbfifo_drain_delay);
3699 	t4_set_reg_field(adap, SGE_INT_ENABLE3,
3700 			 DBFIFO_HP_INT | DBFIFO_LP_INT,
3701 			 DBFIFO_HP_INT | DBFIFO_LP_INT);
3702 	notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3703 }
3704 
3705 static void process_db_drop(struct work_struct *work)
3706 {
3707 	struct adapter *adap;
3708 
3709 	adap = container_of(work, struct adapter, db_drop_task);
3710 
3711 	if (is_t4(adap->chip)) {
3712 		disable_dbs(adap);
3713 		notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3714 		drain_db_fifo(adap, 1);
3715 		recover_all_queues(adap);
3716 		enable_dbs(adap);
3717 	} else {
3718 		u32 dropped_db = t4_read_reg(adap, 0x010ac);
3719 		u16 qid = (dropped_db >> 15) & 0x1ffff;
3720 		u16 pidx_inc = dropped_db & 0x1fff;
3721 		unsigned int s_qpp;
3722 		unsigned short udb_density;
3723 		unsigned long qpshift;
3724 		int page;
3725 		u32 udb;
3726 
3727 		dev_warn(adap->pdev_dev,
3728 			 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
3729 			 dropped_db, qid,
3730 			 (dropped_db >> 14) & 1,
3731 			 (dropped_db >> 13) & 1,
3732 			 pidx_inc);
3733 
3734 		drain_db_fifo(adap, 1);
3735 
3736 		s_qpp = QUEUESPERPAGEPF1 * adap->fn;
3737 		udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
3738 				SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
3739 		qpshift = PAGE_SHIFT - ilog2(udb_density);
3740 		udb = qid << qpshift;
3741 		udb &= PAGE_MASK;
3742 		page = udb / PAGE_SIZE;
3743 		udb += (qid - (page * udb_density)) * 128;
3744 
3745 		writel(PIDX(pidx_inc),  adap->bar2 + udb + 8);
3746 
3747 		/* Re-enable BAR2 WC */
3748 		t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
3749 	}
3750 
3751 	t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3752 }
3753 
3754 void t4_db_full(struct adapter *adap)
3755 {
3756 	if (is_t4(adap->chip)) {
3757 		t4_set_reg_field(adap, SGE_INT_ENABLE3,
3758 				 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3759 		queue_work(workq, &adap->db_full_task);
3760 	}
3761 }
3762 
3763 void t4_db_dropped(struct adapter *adap)
3764 {
3765 	if (is_t4(adap->chip))
3766 		queue_work(workq, &adap->db_drop_task);
3767 }
3768 
3769 static void uld_attach(struct adapter *adap, unsigned int uld)
3770 {
3771 	void *handle;
3772 	struct cxgb4_lld_info lli;
3773 	unsigned short i;
3774 
3775 	lli.pdev = adap->pdev;
3776 	lli.l2t = adap->l2t;
3777 	lli.tids = &adap->tids;
3778 	lli.ports = adap->port;
3779 	lli.vr = &adap->vres;
3780 	lli.mtus = adap->params.mtus;
3781 	if (uld == CXGB4_ULD_RDMA) {
3782 		lli.rxq_ids = adap->sge.rdma_rxq;
3783 		lli.nrxq = adap->sge.rdmaqs;
3784 	} else if (uld == CXGB4_ULD_ISCSI) {
3785 		lli.rxq_ids = adap->sge.ofld_rxq;
3786 		lli.nrxq = adap->sge.ofldqsets;
3787 	}
3788 	lli.ntxq = adap->sge.ofldqsets;
3789 	lli.nchan = adap->params.nports;
3790 	lli.nports = adap->params.nports;
3791 	lli.wr_cred = adap->params.ofldq_wr_cred;
3792 	lli.adapter_type = adap->params.rev;
3793 	lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3794 	lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3795 			t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3796 			(adap->fn * 4));
3797 	lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3798 			t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3799 			(adap->fn * 4));
3800 	lli.filt_mode = adap->filter_mode;
3801 	/* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3802 	for (i = 0; i < NCHAN; i++)
3803 		lli.tx_modq[i] = i;
3804 	lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3805 	lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3806 	lli.fw_vers = adap->params.fw_vers;
3807 	lli.dbfifo_int_thresh = dbfifo_int_thresh;
3808 	lli.sge_pktshift = adap->sge.pktshift;
3809 	lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3810 
3811 	handle = ulds[uld].add(&lli);
3812 	if (IS_ERR(handle)) {
3813 		dev_warn(adap->pdev_dev,
3814 			 "could not attach to the %s driver, error %ld\n",
3815 			 uld_str[uld], PTR_ERR(handle));
3816 		return;
3817 	}
3818 
3819 	adap->uld_handle[uld] = handle;
3820 
3821 	if (!netevent_registered) {
3822 		register_netevent_notifier(&cxgb4_netevent_nb);
3823 		netevent_registered = true;
3824 	}
3825 
3826 	if (adap->flags & FULL_INIT_DONE)
3827 		ulds[uld].state_change(handle, CXGB4_STATE_UP);
3828 }
3829 
3830 static void attach_ulds(struct adapter *adap)
3831 {
3832 	unsigned int i;
3833 
3834 	spin_lock(&adap_rcu_lock);
3835 	list_add_tail_rcu(&adap->rcu_node, &adap_rcu_list);
3836 	spin_unlock(&adap_rcu_lock);
3837 
3838 	mutex_lock(&uld_mutex);
3839 	list_add_tail(&adap->list_node, &adapter_list);
3840 	for (i = 0; i < CXGB4_ULD_MAX; i++)
3841 		if (ulds[i].add)
3842 			uld_attach(adap, i);
3843 	mutex_unlock(&uld_mutex);
3844 }
3845 
3846 static void detach_ulds(struct adapter *adap)
3847 {
3848 	unsigned int i;
3849 
3850 	mutex_lock(&uld_mutex);
3851 	list_del(&adap->list_node);
3852 	for (i = 0; i < CXGB4_ULD_MAX; i++)
3853 		if (adap->uld_handle[i]) {
3854 			ulds[i].state_change(adap->uld_handle[i],
3855 					     CXGB4_STATE_DETACH);
3856 			adap->uld_handle[i] = NULL;
3857 		}
3858 	if (netevent_registered && list_empty(&adapter_list)) {
3859 		unregister_netevent_notifier(&cxgb4_netevent_nb);
3860 		netevent_registered = false;
3861 	}
3862 	mutex_unlock(&uld_mutex);
3863 
3864 	spin_lock(&adap_rcu_lock);
3865 	list_del_rcu(&adap->rcu_node);
3866 	spin_unlock(&adap_rcu_lock);
3867 }
3868 
3869 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3870 {
3871 	unsigned int i;
3872 
3873 	mutex_lock(&uld_mutex);
3874 	for (i = 0; i < CXGB4_ULD_MAX; i++)
3875 		if (adap->uld_handle[i])
3876 			ulds[i].state_change(adap->uld_handle[i], new_state);
3877 	mutex_unlock(&uld_mutex);
3878 }
3879 
3880 /**
3881  *	cxgb4_register_uld - register an upper-layer driver
3882  *	@type: the ULD type
3883  *	@p: the ULD methods
3884  *
3885  *	Registers an upper-layer driver with this driver and notifies the ULD
3886  *	about any presently available devices that support its type.  Returns
3887  *	%-EBUSY if a ULD of the same type is already registered.
3888  */
3889 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3890 {
3891 	int ret = 0;
3892 	struct adapter *adap;
3893 
3894 	if (type >= CXGB4_ULD_MAX)
3895 		return -EINVAL;
3896 	mutex_lock(&uld_mutex);
3897 	if (ulds[type].add) {
3898 		ret = -EBUSY;
3899 		goto out;
3900 	}
3901 	ulds[type] = *p;
3902 	list_for_each_entry(adap, &adapter_list, list_node)
3903 		uld_attach(adap, type);
3904 out:	mutex_unlock(&uld_mutex);
3905 	return ret;
3906 }
3907 EXPORT_SYMBOL(cxgb4_register_uld);
3908 
3909 /**
3910  *	cxgb4_unregister_uld - unregister an upper-layer driver
3911  *	@type: the ULD type
3912  *
3913  *	Unregisters an existing upper-layer driver.
3914  */
3915 int cxgb4_unregister_uld(enum cxgb4_uld type)
3916 {
3917 	struct adapter *adap;
3918 
3919 	if (type >= CXGB4_ULD_MAX)
3920 		return -EINVAL;
3921 	mutex_lock(&uld_mutex);
3922 	list_for_each_entry(adap, &adapter_list, list_node)
3923 		adap->uld_handle[type] = NULL;
3924 	ulds[type].add = NULL;
3925 	mutex_unlock(&uld_mutex);
3926 	return 0;
3927 }
3928 EXPORT_SYMBOL(cxgb4_unregister_uld);
3929 
3930 /* Check if netdev on which event is occured belongs to us or not. Return
3931  * suceess (1) if it belongs otherwise failure (0).
3932  */
3933 static int cxgb4_netdev(struct net_device *netdev)
3934 {
3935 	struct adapter *adap;
3936 	int i;
3937 
3938 	spin_lock(&adap_rcu_lock);
3939 	list_for_each_entry_rcu(adap, &adap_rcu_list, rcu_node)
3940 		for (i = 0; i < MAX_NPORTS; i++)
3941 			if (adap->port[i] == netdev) {
3942 				spin_unlock(&adap_rcu_lock);
3943 				return 1;
3944 			}
3945 	spin_unlock(&adap_rcu_lock);
3946 	return 0;
3947 }
3948 
3949 static int clip_add(struct net_device *event_dev, struct inet6_ifaddr *ifa,
3950 		    unsigned long event)
3951 {
3952 	int ret = NOTIFY_DONE;
3953 
3954 	rcu_read_lock();
3955 	if (cxgb4_netdev(event_dev)) {
3956 		switch (event) {
3957 		case NETDEV_UP:
3958 			ret = cxgb4_clip_get(event_dev,
3959 				(const struct in6_addr *)ifa->addr.s6_addr);
3960 			if (ret < 0) {
3961 				rcu_read_unlock();
3962 				return ret;
3963 			}
3964 			ret = NOTIFY_OK;
3965 			break;
3966 		case NETDEV_DOWN:
3967 			cxgb4_clip_release(event_dev,
3968 				(const struct in6_addr *)ifa->addr.s6_addr);
3969 			ret = NOTIFY_OK;
3970 			break;
3971 		default:
3972 			break;
3973 		}
3974 	}
3975 	rcu_read_unlock();
3976 	return ret;
3977 }
3978 
3979 static int cxgb4_inet6addr_handler(struct notifier_block *this,
3980 		unsigned long event, void *data)
3981 {
3982 	struct inet6_ifaddr *ifa = data;
3983 	struct net_device *event_dev;
3984 	int ret = NOTIFY_DONE;
3985 	struct bonding *bond = netdev_priv(ifa->idev->dev);
3986 	struct slave *slave;
3987 	struct pci_dev *first_pdev = NULL;
3988 
3989 	if (ifa->idev->dev->priv_flags & IFF_802_1Q_VLAN) {
3990 		event_dev = vlan_dev_real_dev(ifa->idev->dev);
3991 		ret = clip_add(event_dev, ifa, event);
3992 	} else if (ifa->idev->dev->flags & IFF_MASTER) {
3993 		/* It is possible that two different adapters are bonded in one
3994 		 * bond. We need to find such different adapters and add clip
3995 		 * in all of them only once.
3996 		 */
3997 		read_lock(&bond->lock);
3998 		bond_for_each_slave(bond, slave) {
3999 			if (!first_pdev) {
4000 				ret = clip_add(slave->dev, ifa, event);
4001 				/* If clip_add is success then only initialize
4002 				 * first_pdev since it means it is our device
4003 				 */
4004 				if (ret == NOTIFY_OK)
4005 					first_pdev = to_pci_dev(
4006 							slave->dev->dev.parent);
4007 			} else if (first_pdev !=
4008 				   to_pci_dev(slave->dev->dev.parent))
4009 					ret = clip_add(slave->dev, ifa, event);
4010 		}
4011 		read_unlock(&bond->lock);
4012 	} else
4013 		ret = clip_add(ifa->idev->dev, ifa, event);
4014 
4015 	return ret;
4016 }
4017 
4018 static struct notifier_block cxgb4_inet6addr_notifier = {
4019 	.notifier_call = cxgb4_inet6addr_handler
4020 };
4021 
4022 /* Retrieves IPv6 addresses from a root device (bond, vlan) associated with
4023  * a physical device.
4024  * The physical device reference is needed to send the actul CLIP command.
4025  */
4026 static int update_dev_clip(struct net_device *root_dev, struct net_device *dev)
4027 {
4028 	struct inet6_dev *idev = NULL;
4029 	struct inet6_ifaddr *ifa;
4030 	int ret = 0;
4031 
4032 	idev = __in6_dev_get(root_dev);
4033 	if (!idev)
4034 		return ret;
4035 
4036 	read_lock_bh(&idev->lock);
4037 	list_for_each_entry(ifa, &idev->addr_list, if_list) {
4038 		ret = cxgb4_clip_get(dev,
4039 				(const struct in6_addr *)ifa->addr.s6_addr);
4040 		if (ret < 0)
4041 			break;
4042 	}
4043 	read_unlock_bh(&idev->lock);
4044 
4045 	return ret;
4046 }
4047 
4048 static int update_root_dev_clip(struct net_device *dev)
4049 {
4050 	struct net_device *root_dev = NULL;
4051 	int i, ret = 0;
4052 
4053 	/* First populate the real net device's IPv6 addresses */
4054 	ret = update_dev_clip(dev, dev);
4055 	if (ret)
4056 		return ret;
4057 
4058 	/* Parse all bond and vlan devices layered on top of the physical dev */
4059 	for (i = 0; i < VLAN_N_VID; i++) {
4060 		root_dev = __vlan_find_dev_deep(dev, htons(ETH_P_8021Q), i);
4061 		if (!root_dev)
4062 			continue;
4063 
4064 		ret = update_dev_clip(root_dev, dev);
4065 		if (ret)
4066 			break;
4067 	}
4068 	return ret;
4069 }
4070 
4071 static void update_clip(const struct adapter *adap)
4072 {
4073 	int i;
4074 	struct net_device *dev;
4075 	int ret;
4076 
4077 	rcu_read_lock();
4078 
4079 	for (i = 0; i < MAX_NPORTS; i++) {
4080 		dev = adap->port[i];
4081 		ret = 0;
4082 
4083 		if (dev)
4084 			ret = update_root_dev_clip(dev);
4085 
4086 		if (ret < 0)
4087 			break;
4088 	}
4089 	rcu_read_unlock();
4090 }
4091 
4092 /**
4093  *	cxgb_up - enable the adapter
4094  *	@adap: adapter being enabled
4095  *
4096  *	Called when the first port is enabled, this function performs the
4097  *	actions necessary to make an adapter operational, such as completing
4098  *	the initialization of HW modules, and enabling interrupts.
4099  *
4100  *	Must be called with the rtnl lock held.
4101  */
4102 static int cxgb_up(struct adapter *adap)
4103 {
4104 	int err;
4105 
4106 	err = setup_sge_queues(adap);
4107 	if (err)
4108 		goto out;
4109 	err = setup_rss(adap);
4110 	if (err)
4111 		goto freeq;
4112 
4113 	if (adap->flags & USING_MSIX) {
4114 		name_msix_vecs(adap);
4115 		err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
4116 				  adap->msix_info[0].desc, adap);
4117 		if (err)
4118 			goto irq_err;
4119 
4120 		err = request_msix_queue_irqs(adap);
4121 		if (err) {
4122 			free_irq(adap->msix_info[0].vec, adap);
4123 			goto irq_err;
4124 		}
4125 	} else {
4126 		err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
4127 				  (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
4128 				  adap->port[0]->name, adap);
4129 		if (err)
4130 			goto irq_err;
4131 	}
4132 	enable_rx(adap);
4133 	t4_sge_start(adap);
4134 	t4_intr_enable(adap);
4135 	adap->flags |= FULL_INIT_DONE;
4136 	notify_ulds(adap, CXGB4_STATE_UP);
4137 	update_clip(adap);
4138  out:
4139 	return err;
4140  irq_err:
4141 	dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
4142  freeq:
4143 	t4_free_sge_resources(adap);
4144 	goto out;
4145 }
4146 
4147 static void cxgb_down(struct adapter *adapter)
4148 {
4149 	t4_intr_disable(adapter);
4150 	cancel_work_sync(&adapter->tid_release_task);
4151 	cancel_work_sync(&adapter->db_full_task);
4152 	cancel_work_sync(&adapter->db_drop_task);
4153 	adapter->tid_release_task_busy = false;
4154 	adapter->tid_release_head = NULL;
4155 
4156 	if (adapter->flags & USING_MSIX) {
4157 		free_msix_queue_irqs(adapter);
4158 		free_irq(adapter->msix_info[0].vec, adapter);
4159 	} else
4160 		free_irq(adapter->pdev->irq, adapter);
4161 	quiesce_rx(adapter);
4162 	t4_sge_stop(adapter);
4163 	t4_free_sge_resources(adapter);
4164 	adapter->flags &= ~FULL_INIT_DONE;
4165 }
4166 
4167 /*
4168  * net_device operations
4169  */
4170 static int cxgb_open(struct net_device *dev)
4171 {
4172 	int err;
4173 	struct port_info *pi = netdev_priv(dev);
4174 	struct adapter *adapter = pi->adapter;
4175 
4176 	netif_carrier_off(dev);
4177 
4178 	if (!(adapter->flags & FULL_INIT_DONE)) {
4179 		err = cxgb_up(adapter);
4180 		if (err < 0)
4181 			return err;
4182 	}
4183 
4184 	err = link_start(dev);
4185 	if (!err)
4186 		netif_tx_start_all_queues(dev);
4187 	return err;
4188 }
4189 
4190 static int cxgb_close(struct net_device *dev)
4191 {
4192 	struct port_info *pi = netdev_priv(dev);
4193 	struct adapter *adapter = pi->adapter;
4194 
4195 	netif_tx_stop_all_queues(dev);
4196 	netif_carrier_off(dev);
4197 	return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
4198 }
4199 
4200 /* Return an error number if the indicated filter isn't writable ...
4201  */
4202 static int writable_filter(struct filter_entry *f)
4203 {
4204 	if (f->locked)
4205 		return -EPERM;
4206 	if (f->pending)
4207 		return -EBUSY;
4208 
4209 	return 0;
4210 }
4211 
4212 /* Delete the filter at the specified index (if valid).  The checks for all
4213  * the common problems with doing this like the filter being locked, currently
4214  * pending in another operation, etc.
4215  */
4216 static int delete_filter(struct adapter *adapter, unsigned int fidx)
4217 {
4218 	struct filter_entry *f;
4219 	int ret;
4220 
4221 	if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
4222 		return -EINVAL;
4223 
4224 	f = &adapter->tids.ftid_tab[fidx];
4225 	ret = writable_filter(f);
4226 	if (ret)
4227 		return ret;
4228 	if (f->valid)
4229 		return del_filter_wr(adapter, fidx);
4230 
4231 	return 0;
4232 }
4233 
4234 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
4235 		__be32 sip, __be16 sport, __be16 vlan,
4236 		unsigned int queue, unsigned char port, unsigned char mask)
4237 {
4238 	int ret;
4239 	struct filter_entry *f;
4240 	struct adapter *adap;
4241 	int i;
4242 	u8 *val;
4243 
4244 	adap = netdev2adap(dev);
4245 
4246 	/* Adjust stid to correct filter index */
4247 	stid -= adap->tids.nstids;
4248 	stid += adap->tids.nftids;
4249 
4250 	/* Check to make sure the filter requested is writable ...
4251 	 */
4252 	f = &adap->tids.ftid_tab[stid];
4253 	ret = writable_filter(f);
4254 	if (ret)
4255 		return ret;
4256 
4257 	/* Clear out any old resources being used by the filter before
4258 	 * we start constructing the new filter.
4259 	 */
4260 	if (f->valid)
4261 		clear_filter(adap, f);
4262 
4263 	/* Clear out filter specifications */
4264 	memset(&f->fs, 0, sizeof(struct ch_filter_specification));
4265 	f->fs.val.lport = cpu_to_be16(sport);
4266 	f->fs.mask.lport  = ~0;
4267 	val = (u8 *)&sip;
4268 	if ((val[0] | val[1] | val[2] | val[3]) != 0) {
4269 		for (i = 0; i < 4; i++) {
4270 			f->fs.val.lip[i] = val[i];
4271 			f->fs.mask.lip[i] = ~0;
4272 		}
4273 		if (adap->filter_mode & F_PORT) {
4274 			f->fs.val.iport = port;
4275 			f->fs.mask.iport = mask;
4276 		}
4277 	}
4278 
4279 	f->fs.dirsteer = 1;
4280 	f->fs.iq = queue;
4281 	/* Mark filter as locked */
4282 	f->locked = 1;
4283 	f->fs.rpttid = 1;
4284 
4285 	ret = set_filter_wr(adap, stid);
4286 	if (ret) {
4287 		clear_filter(adap, f);
4288 		return ret;
4289 	}
4290 
4291 	return 0;
4292 }
4293 EXPORT_SYMBOL(cxgb4_create_server_filter);
4294 
4295 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
4296 		unsigned int queue, bool ipv6)
4297 {
4298 	int ret;
4299 	struct filter_entry *f;
4300 	struct adapter *adap;
4301 
4302 	adap = netdev2adap(dev);
4303 
4304 	/* Adjust stid to correct filter index */
4305 	stid -= adap->tids.nstids;
4306 	stid += adap->tids.nftids;
4307 
4308 	f = &adap->tids.ftid_tab[stid];
4309 	/* Unlock the filter */
4310 	f->locked = 0;
4311 
4312 	ret = delete_filter(adap, stid);
4313 	if (ret)
4314 		return ret;
4315 
4316 	return 0;
4317 }
4318 EXPORT_SYMBOL(cxgb4_remove_server_filter);
4319 
4320 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
4321 						struct rtnl_link_stats64 *ns)
4322 {
4323 	struct port_stats stats;
4324 	struct port_info *p = netdev_priv(dev);
4325 	struct adapter *adapter = p->adapter;
4326 
4327 	spin_lock(&adapter->stats_lock);
4328 	t4_get_port_stats(adapter, p->tx_chan, &stats);
4329 	spin_unlock(&adapter->stats_lock);
4330 
4331 	ns->tx_bytes   = stats.tx_octets;
4332 	ns->tx_packets = stats.tx_frames;
4333 	ns->rx_bytes   = stats.rx_octets;
4334 	ns->rx_packets = stats.rx_frames;
4335 	ns->multicast  = stats.rx_mcast_frames;
4336 
4337 	/* detailed rx_errors */
4338 	ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
4339 			       stats.rx_runt;
4340 	ns->rx_over_errors   = 0;
4341 	ns->rx_crc_errors    = stats.rx_fcs_err;
4342 	ns->rx_frame_errors  = stats.rx_symbol_err;
4343 	ns->rx_fifo_errors   = stats.rx_ovflow0 + stats.rx_ovflow1 +
4344 			       stats.rx_ovflow2 + stats.rx_ovflow3 +
4345 			       stats.rx_trunc0 + stats.rx_trunc1 +
4346 			       stats.rx_trunc2 + stats.rx_trunc3;
4347 	ns->rx_missed_errors = 0;
4348 
4349 	/* detailed tx_errors */
4350 	ns->tx_aborted_errors   = 0;
4351 	ns->tx_carrier_errors   = 0;
4352 	ns->tx_fifo_errors      = 0;
4353 	ns->tx_heartbeat_errors = 0;
4354 	ns->tx_window_errors    = 0;
4355 
4356 	ns->tx_errors = stats.tx_error_frames;
4357 	ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
4358 		ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
4359 	return ns;
4360 }
4361 
4362 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
4363 {
4364 	unsigned int mbox;
4365 	int ret = 0, prtad, devad;
4366 	struct port_info *pi = netdev_priv(dev);
4367 	struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
4368 
4369 	switch (cmd) {
4370 	case SIOCGMIIPHY:
4371 		if (pi->mdio_addr < 0)
4372 			return -EOPNOTSUPP;
4373 		data->phy_id = pi->mdio_addr;
4374 		break;
4375 	case SIOCGMIIREG:
4376 	case SIOCSMIIREG:
4377 		if (mdio_phy_id_is_c45(data->phy_id)) {
4378 			prtad = mdio_phy_id_prtad(data->phy_id);
4379 			devad = mdio_phy_id_devad(data->phy_id);
4380 		} else if (data->phy_id < 32) {
4381 			prtad = data->phy_id;
4382 			devad = 0;
4383 			data->reg_num &= 0x1f;
4384 		} else
4385 			return -EINVAL;
4386 
4387 		mbox = pi->adapter->fn;
4388 		if (cmd == SIOCGMIIREG)
4389 			ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
4390 					 data->reg_num, &data->val_out);
4391 		else
4392 			ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
4393 					 data->reg_num, data->val_in);
4394 		break;
4395 	default:
4396 		return -EOPNOTSUPP;
4397 	}
4398 	return ret;
4399 }
4400 
4401 static void cxgb_set_rxmode(struct net_device *dev)
4402 {
4403 	/* unfortunately we can't return errors to the stack */
4404 	set_rxmode(dev, -1, false);
4405 }
4406 
4407 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
4408 {
4409 	int ret;
4410 	struct port_info *pi = netdev_priv(dev);
4411 
4412 	if (new_mtu < 81 || new_mtu > MAX_MTU)         /* accommodate SACK */
4413 		return -EINVAL;
4414 	ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
4415 			    -1, -1, -1, true);
4416 	if (!ret)
4417 		dev->mtu = new_mtu;
4418 	return ret;
4419 }
4420 
4421 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
4422 {
4423 	int ret;
4424 	struct sockaddr *addr = p;
4425 	struct port_info *pi = netdev_priv(dev);
4426 
4427 	if (!is_valid_ether_addr(addr->sa_data))
4428 		return -EADDRNOTAVAIL;
4429 
4430 	ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
4431 			    pi->xact_addr_filt, addr->sa_data, true, true);
4432 	if (ret < 0)
4433 		return ret;
4434 
4435 	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4436 	pi->xact_addr_filt = ret;
4437 	return 0;
4438 }
4439 
4440 #ifdef CONFIG_NET_POLL_CONTROLLER
4441 static void cxgb_netpoll(struct net_device *dev)
4442 {
4443 	struct port_info *pi = netdev_priv(dev);
4444 	struct adapter *adap = pi->adapter;
4445 
4446 	if (adap->flags & USING_MSIX) {
4447 		int i;
4448 		struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4449 
4450 		for (i = pi->nqsets; i; i--, rx++)
4451 			t4_sge_intr_msix(0, &rx->rspq);
4452 	} else
4453 		t4_intr_handler(adap)(0, adap);
4454 }
4455 #endif
4456 
4457 static const struct net_device_ops cxgb4_netdev_ops = {
4458 	.ndo_open             = cxgb_open,
4459 	.ndo_stop             = cxgb_close,
4460 	.ndo_start_xmit       = t4_eth_xmit,
4461 	.ndo_get_stats64      = cxgb_get_stats,
4462 	.ndo_set_rx_mode      = cxgb_set_rxmode,
4463 	.ndo_set_mac_address  = cxgb_set_mac_addr,
4464 	.ndo_set_features     = cxgb_set_features,
4465 	.ndo_validate_addr    = eth_validate_addr,
4466 	.ndo_do_ioctl         = cxgb_ioctl,
4467 	.ndo_change_mtu       = cxgb_change_mtu,
4468 #ifdef CONFIG_NET_POLL_CONTROLLER
4469 	.ndo_poll_controller  = cxgb_netpoll,
4470 #endif
4471 };
4472 
4473 void t4_fatal_err(struct adapter *adap)
4474 {
4475 	t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4476 	t4_intr_disable(adap);
4477 	dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4478 }
4479 
4480 static void setup_memwin(struct adapter *adap)
4481 {
4482 	u32 bar0, mem_win0_base, mem_win1_base, mem_win2_base;
4483 
4484 	bar0 = pci_resource_start(adap->pdev, 0);  /* truncation intentional */
4485 	if (is_t4(adap->chip)) {
4486 		mem_win0_base = bar0 + MEMWIN0_BASE;
4487 		mem_win1_base = bar0 + MEMWIN1_BASE;
4488 		mem_win2_base = bar0 + MEMWIN2_BASE;
4489 	} else {
4490 		/* For T5, only relative offset inside the PCIe BAR is passed */
4491 		mem_win0_base = MEMWIN0_BASE;
4492 		mem_win1_base = MEMWIN1_BASE_T5;
4493 		mem_win2_base = MEMWIN2_BASE_T5;
4494 	}
4495 	t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4496 		     mem_win0_base | BIR(0) |
4497 		     WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4498 	t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4499 		     mem_win1_base | BIR(0) |
4500 		     WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4501 	t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4502 		     mem_win2_base | BIR(0) |
4503 		     WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
4504 }
4505 
4506 static void setup_memwin_rdma(struct adapter *adap)
4507 {
4508 	if (adap->vres.ocq.size) {
4509 		unsigned int start, sz_kb;
4510 
4511 		start = pci_resource_start(adap->pdev, 2) +
4512 			OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4513 		sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4514 		t4_write_reg(adap,
4515 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4516 			     start | BIR(1) | WINDOW(ilog2(sz_kb)));
4517 		t4_write_reg(adap,
4518 			     PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4519 			     adap->vres.ocq.start);
4520 		t4_read_reg(adap,
4521 			    PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4522 	}
4523 }
4524 
4525 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4526 {
4527 	u32 v;
4528 	int ret;
4529 
4530 	/* get device capabilities */
4531 	memset(c, 0, sizeof(*c));
4532 	c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4533 			       FW_CMD_REQUEST | FW_CMD_READ);
4534 	c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4535 	ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4536 	if (ret < 0)
4537 		return ret;
4538 
4539 	/* select capabilities we'll be using */
4540 	if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4541 		if (!vf_acls)
4542 			c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4543 		else
4544 			c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4545 	} else if (vf_acls) {
4546 		dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4547 		return ret;
4548 	}
4549 	c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4550 			       FW_CMD_REQUEST | FW_CMD_WRITE);
4551 	ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4552 	if (ret < 0)
4553 		return ret;
4554 
4555 	ret = t4_config_glbl_rss(adap, adap->fn,
4556 				 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4557 				 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4558 				 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4559 	if (ret < 0)
4560 		return ret;
4561 
4562 	ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4563 			  0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4564 	if (ret < 0)
4565 		return ret;
4566 
4567 	t4_sge_init(adap);
4568 
4569 	/* tweak some settings */
4570 	t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4571 	t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4572 	t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4573 	v = t4_read_reg(adap, TP_PIO_DATA);
4574 	t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4575 
4576 	/* first 4 Tx modulation queues point to consecutive Tx channels */
4577 	adap->params.tp.tx_modq_map = 0xE4;
4578 	t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4579 		     V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4580 
4581 	/* associate each Tx modulation queue with consecutive Tx channels */
4582 	v = 0x84218421;
4583 	t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4584 			  &v, 1, A_TP_TX_SCHED_HDR);
4585 	t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4586 			  &v, 1, A_TP_TX_SCHED_FIFO);
4587 	t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4588 			  &v, 1, A_TP_TX_SCHED_PCMD);
4589 
4590 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4591 	if (is_offload(adap)) {
4592 		t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4593 			     V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4594 			     V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4595 			     V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4596 			     V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4597 		t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4598 			     V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4599 			     V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4600 			     V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4601 			     V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4602 	}
4603 
4604 	/* get basic stuff going */
4605 	return t4_early_init(adap, adap->fn);
4606 }
4607 
4608 /*
4609  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
4610  */
4611 #define MAX_ATIDS 8192U
4612 
4613 /*
4614  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4615  *
4616  * If the firmware we're dealing with has Configuration File support, then
4617  * we use that to perform all configuration
4618  */
4619 
4620 /*
4621  * Tweak configuration based on module parameters, etc.  Most of these have
4622  * defaults assigned to them by Firmware Configuration Files (if we're using
4623  * them) but need to be explicitly set if we're using hard-coded
4624  * initialization.  But even in the case of using Firmware Configuration
4625  * Files, we'd like to expose the ability to change these via module
4626  * parameters so these are essentially common tweaks/settings for
4627  * Configuration Files and hard-coded initialization ...
4628  */
4629 static int adap_init0_tweaks(struct adapter *adapter)
4630 {
4631 	/*
4632 	 * Fix up various Host-Dependent Parameters like Page Size, Cache
4633 	 * Line Size, etc.  The firmware default is for a 4KB Page Size and
4634 	 * 64B Cache Line Size ...
4635 	 */
4636 	t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4637 
4638 	/*
4639 	 * Process module parameters which affect early initialization.
4640 	 */
4641 	if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4642 		dev_err(&adapter->pdev->dev,
4643 			"Ignoring illegal rx_dma_offset=%d, using 2\n",
4644 			rx_dma_offset);
4645 		rx_dma_offset = 2;
4646 	}
4647 	t4_set_reg_field(adapter, SGE_CONTROL,
4648 			 PKTSHIFT_MASK,
4649 			 PKTSHIFT(rx_dma_offset));
4650 
4651 	/*
4652 	 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4653 	 * adds the pseudo header itself.
4654 	 */
4655 	t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
4656 			       CSUM_HAS_PSEUDO_HDR, 0);
4657 
4658 	return 0;
4659 }
4660 
4661 /*
4662  * Attempt to initialize the adapter via a Firmware Configuration File.
4663  */
4664 static int adap_init0_config(struct adapter *adapter, int reset)
4665 {
4666 	struct fw_caps_config_cmd caps_cmd;
4667 	const struct firmware *cf;
4668 	unsigned long mtype = 0, maddr = 0;
4669 	u32 finiver, finicsum, cfcsum;
4670 	int ret, using_flash;
4671 	char *fw_config_file, fw_config_file_path[256];
4672 
4673 	/*
4674 	 * Reset device if necessary.
4675 	 */
4676 	if (reset) {
4677 		ret = t4_fw_reset(adapter, adapter->mbox,
4678 				  PIORSTMODE | PIORST);
4679 		if (ret < 0)
4680 			goto bye;
4681 	}
4682 
4683 	/*
4684 	 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4685 	 * then use that.  Otherwise, use the configuration file stored
4686 	 * in the adapter flash ...
4687 	 */
4688 	switch (CHELSIO_CHIP_VERSION(adapter->chip)) {
4689 	case CHELSIO_T4:
4690 		fw_config_file = FW_CFNAME;
4691 		break;
4692 	case CHELSIO_T5:
4693 		fw_config_file = FW5_CFNAME;
4694 		break;
4695 	default:
4696 		dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4697 		       adapter->pdev->device);
4698 		ret = -EINVAL;
4699 		goto bye;
4700 	}
4701 
4702 	ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4703 	if (ret < 0) {
4704 		using_flash = 1;
4705 		mtype = FW_MEMTYPE_CF_FLASH;
4706 		maddr = t4_flash_cfg_addr(adapter);
4707 	} else {
4708 		u32 params[7], val[7];
4709 
4710 		using_flash = 0;
4711 		if (cf->size >= FLASH_CFG_MAX_SIZE)
4712 			ret = -ENOMEM;
4713 		else {
4714 			params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4715 			     FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4716 			ret = t4_query_params(adapter, adapter->mbox,
4717 					      adapter->fn, 0, 1, params, val);
4718 			if (ret == 0) {
4719 				/*
4720 				 * For t4_memory_write() below addresses and
4721 				 * sizes have to be in terms of multiples of 4
4722 				 * bytes.  So, if the Configuration File isn't
4723 				 * a multiple of 4 bytes in length we'll have
4724 				 * to write that out separately since we can't
4725 				 * guarantee that the bytes following the
4726 				 * residual byte in the buffer returned by
4727 				 * request_firmware() are zeroed out ...
4728 				 */
4729 				size_t resid = cf->size & 0x3;
4730 				size_t size = cf->size & ~0x3;
4731 				__be32 *data = (__be32 *)cf->data;
4732 
4733 				mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
4734 				maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
4735 
4736 				ret = t4_memory_write(adapter, mtype, maddr,
4737 						      size, data);
4738 				if (ret == 0 && resid != 0) {
4739 					union {
4740 						__be32 word;
4741 						char buf[4];
4742 					} last;
4743 					int i;
4744 
4745 					last.word = data[size >> 2];
4746 					for (i = resid; i < 4; i++)
4747 						last.buf[i] = 0;
4748 					ret = t4_memory_write(adapter, mtype,
4749 							      maddr + size,
4750 							      4, &last.word);
4751 				}
4752 			}
4753 		}
4754 
4755 		release_firmware(cf);
4756 		if (ret)
4757 			goto bye;
4758 	}
4759 
4760 	/*
4761 	 * Issue a Capability Configuration command to the firmware to get it
4762 	 * to parse the Configuration File.  We don't use t4_fw_config_file()
4763 	 * because we want the ability to modify various features after we've
4764 	 * processed the configuration file ...
4765 	 */
4766 	memset(&caps_cmd, 0, sizeof(caps_cmd));
4767 	caps_cmd.op_to_write =
4768 		htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4769 		      FW_CMD_REQUEST |
4770 		      FW_CMD_READ);
4771 	caps_cmd.cfvalid_to_len16 =
4772 		htonl(FW_CAPS_CONFIG_CMD_CFVALID |
4773 		      FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4774 		      FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
4775 		      FW_LEN16(caps_cmd));
4776 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4777 			 &caps_cmd);
4778 	if (ret < 0)
4779 		goto bye;
4780 
4781 	finiver = ntohl(caps_cmd.finiver);
4782 	finicsum = ntohl(caps_cmd.finicsum);
4783 	cfcsum = ntohl(caps_cmd.cfcsum);
4784 	if (finicsum != cfcsum)
4785 		dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4786 			 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4787 			 finicsum, cfcsum);
4788 
4789 	/*
4790 	 * And now tell the firmware to use the configuration we just loaded.
4791 	 */
4792 	caps_cmd.op_to_write =
4793 		htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4794 		      FW_CMD_REQUEST |
4795 		      FW_CMD_WRITE);
4796 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4797 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4798 			 NULL);
4799 	if (ret < 0)
4800 		goto bye;
4801 
4802 	/*
4803 	 * Tweak configuration based on system architecture, module
4804 	 * parameters, etc.
4805 	 */
4806 	ret = adap_init0_tweaks(adapter);
4807 	if (ret < 0)
4808 		goto bye;
4809 
4810 	/*
4811 	 * And finally tell the firmware to initialize itself using the
4812 	 * parameters from the Configuration File.
4813 	 */
4814 	ret = t4_fw_initialize(adapter, adapter->mbox);
4815 	if (ret < 0)
4816 		goto bye;
4817 
4818 	sprintf(fw_config_file_path, "/lib/firmware/%s", fw_config_file);
4819 	/*
4820 	 * Return successfully and note that we're operating with parameters
4821 	 * not supplied by the driver, rather than from hard-wired
4822 	 * initialization constants burried in the driver.
4823 	 */
4824 	adapter->flags |= USING_SOFT_PARAMS;
4825 	dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4826 		 "Configuration File %s, version %#x, computed checksum %#x\n",
4827 		 (using_flash
4828 		  ? "in device FLASH"
4829 		  : fw_config_file_path),
4830 		 finiver, cfcsum);
4831 	return 0;
4832 
4833 	/*
4834 	 * Something bad happened.  Return the error ...  (If the "error"
4835 	 * is that there's no Configuration File on the adapter we don't
4836 	 * want to issue a warning since this is fairly common.)
4837 	 */
4838 bye:
4839 	if (ret != -ENOENT)
4840 		dev_warn(adapter->pdev_dev, "Configuration file error %d\n",
4841 			 -ret);
4842 	return ret;
4843 }
4844 
4845 /*
4846  * Attempt to initialize the adapter via hard-coded, driver supplied
4847  * parameters ...
4848  */
4849 static int adap_init0_no_config(struct adapter *adapter, int reset)
4850 {
4851 	struct sge *s = &adapter->sge;
4852 	struct fw_caps_config_cmd caps_cmd;
4853 	u32 v;
4854 	int i, ret;
4855 
4856 	/*
4857 	 * Reset device if necessary
4858 	 */
4859 	if (reset) {
4860 		ret = t4_fw_reset(adapter, adapter->mbox,
4861 				  PIORSTMODE | PIORST);
4862 		if (ret < 0)
4863 			goto bye;
4864 	}
4865 
4866 	/*
4867 	 * Get device capabilities and select which we'll be using.
4868 	 */
4869 	memset(&caps_cmd, 0, sizeof(caps_cmd));
4870 	caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4871 				     FW_CMD_REQUEST | FW_CMD_READ);
4872 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4873 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4874 			 &caps_cmd);
4875 	if (ret < 0)
4876 		goto bye;
4877 
4878 	if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4879 		if (!vf_acls)
4880 			caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4881 		else
4882 			caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4883 	} else if (vf_acls) {
4884 		dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
4885 		goto bye;
4886 	}
4887 	caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4888 			      FW_CMD_REQUEST | FW_CMD_WRITE);
4889 	ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4890 			 NULL);
4891 	if (ret < 0)
4892 		goto bye;
4893 
4894 	/*
4895 	 * Tweak configuration based on system architecture, module
4896 	 * parameters, etc.
4897 	 */
4898 	ret = adap_init0_tweaks(adapter);
4899 	if (ret < 0)
4900 		goto bye;
4901 
4902 	/*
4903 	 * Select RSS Global Mode we want to use.  We use "Basic Virtual"
4904 	 * mode which maps each Virtual Interface to its own section of
4905 	 * the RSS Table and we turn on all map and hash enables ...
4906 	 */
4907 	adapter->flags |= RSS_TNLALLLOOKUP;
4908 	ret = t4_config_glbl_rss(adapter, adapter->mbox,
4909 				 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4910 				 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4911 				 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4912 				 ((adapter->flags & RSS_TNLALLLOOKUP) ?
4913 					FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4914 	if (ret < 0)
4915 		goto bye;
4916 
4917 	/*
4918 	 * Set up our own fundamental resource provisioning ...
4919 	 */
4920 	ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4921 			  PFRES_NEQ, PFRES_NETHCTRL,
4922 			  PFRES_NIQFLINT, PFRES_NIQ,
4923 			  PFRES_TC, PFRES_NVI,
4924 			  FW_PFVF_CMD_CMASK_MASK,
4925 			  pfvfres_pmask(adapter, adapter->fn, 0),
4926 			  PFRES_NEXACTF,
4927 			  PFRES_R_CAPS, PFRES_WX_CAPS);
4928 	if (ret < 0)
4929 		goto bye;
4930 
4931 	/*
4932 	 * Perform low level SGE initialization.  We need to do this before we
4933 	 * send the firmware the INITIALIZE command because that will cause
4934 	 * any other PF Drivers which are waiting for the Master
4935 	 * Initialization to proceed forward.
4936 	 */
4937 	for (i = 0; i < SGE_NTIMERS - 1; i++)
4938 		s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4939 	s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4940 	s->counter_val[0] = 1;
4941 	for (i = 1; i < SGE_NCOUNTERS; i++)
4942 		s->counter_val[i] = min(intr_cnt[i - 1],
4943 					THRESHOLD_0_GET(THRESHOLD_0_MASK));
4944 	t4_sge_init(adapter);
4945 
4946 #ifdef CONFIG_PCI_IOV
4947 	/*
4948 	 * Provision resource limits for Virtual Functions.  We currently
4949 	 * grant them all the same static resource limits except for the Port
4950 	 * Access Rights Mask which we're assigning based on the PF.  All of
4951 	 * the static provisioning stuff for both the PF and VF really needs
4952 	 * to be managed in a persistent manner for each device which the
4953 	 * firmware controls.
4954 	 */
4955 	{
4956 		int pf, vf;
4957 
4958 		for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
4959 			if (num_vf[pf] <= 0)
4960 				continue;
4961 
4962 			/* VF numbering starts at 1! */
4963 			for (vf = 1; vf <= num_vf[pf]; vf++) {
4964 				ret = t4_cfg_pfvf(adapter, adapter->mbox,
4965 						  pf, vf,
4966 						  VFRES_NEQ, VFRES_NETHCTRL,
4967 						  VFRES_NIQFLINT, VFRES_NIQ,
4968 						  VFRES_TC, VFRES_NVI,
4969 						  FW_PFVF_CMD_CMASK_MASK,
4970 						  pfvfres_pmask(
4971 						  adapter, pf, vf),
4972 						  VFRES_NEXACTF,
4973 						  VFRES_R_CAPS, VFRES_WX_CAPS);
4974 				if (ret < 0)
4975 					dev_warn(adapter->pdev_dev,
4976 						 "failed to "\
4977 						 "provision pf/vf=%d/%d; "
4978 						 "err=%d\n", pf, vf, ret);
4979 			}
4980 		}
4981 	}
4982 #endif
4983 
4984 	/*
4985 	 * Set up the default filter mode.  Later we'll want to implement this
4986 	 * via a firmware command, etc. ...  This needs to be done before the
4987 	 * firmare initialization command ...  If the selected set of fields
4988 	 * isn't equal to the default value, we'll need to make sure that the
4989 	 * field selections will fit in the 36-bit budget.
4990 	 */
4991 	if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
4992 		int j, bits = 0;
4993 
4994 		for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
4995 			switch (tp_vlan_pri_map & (1 << j)) {
4996 			case 0:
4997 				/* compressed filter field not enabled */
4998 				break;
4999 			case FCOE_MASK:
5000 				bits +=  1;
5001 				break;
5002 			case PORT_MASK:
5003 				bits +=  3;
5004 				break;
5005 			case VNIC_ID_MASK:
5006 				bits += 17;
5007 				break;
5008 			case VLAN_MASK:
5009 				bits += 17;
5010 				break;
5011 			case TOS_MASK:
5012 				bits +=  8;
5013 				break;
5014 			case PROTOCOL_MASK:
5015 				bits +=  8;
5016 				break;
5017 			case ETHERTYPE_MASK:
5018 				bits += 16;
5019 				break;
5020 			case MACMATCH_MASK:
5021 				bits +=  9;
5022 				break;
5023 			case MPSHITTYPE_MASK:
5024 				bits +=  3;
5025 				break;
5026 			case FRAGMENTATION_MASK:
5027 				bits +=  1;
5028 				break;
5029 			}
5030 
5031 		if (bits > 36) {
5032 			dev_err(adapter->pdev_dev,
5033 				"tp_vlan_pri_map=%#x needs %d bits > 36;"\
5034 				" using %#x\n", tp_vlan_pri_map, bits,
5035 				TP_VLAN_PRI_MAP_DEFAULT);
5036 			tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
5037 		}
5038 	}
5039 	v = tp_vlan_pri_map;
5040 	t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
5041 			  &v, 1, TP_VLAN_PRI_MAP);
5042 
5043 	/*
5044 	 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
5045 	 * to support any of the compressed filter fields above.  Newer
5046 	 * versions of the firmware do this automatically but it doesn't hurt
5047 	 * to set it here.  Meanwhile, we do _not_ need to set Lookup Every
5048 	 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
5049 	 * since the firmware automatically turns this on and off when we have
5050 	 * a non-zero number of filters active (since it does have a
5051 	 * performance impact).
5052 	 */
5053 	if (tp_vlan_pri_map)
5054 		t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
5055 				 FIVETUPLELOOKUP_MASK,
5056 				 FIVETUPLELOOKUP_MASK);
5057 
5058 	/*
5059 	 * Tweak some settings.
5060 	 */
5061 	t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
5062 		     RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
5063 		     PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
5064 		     KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
5065 
5066 	/*
5067 	 * Get basic stuff going by issuing the Firmware Initialize command.
5068 	 * Note that this _must_ be after all PFVF commands ...
5069 	 */
5070 	ret = t4_fw_initialize(adapter, adapter->mbox);
5071 	if (ret < 0)
5072 		goto bye;
5073 
5074 	/*
5075 	 * Return successfully!
5076 	 */
5077 	dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
5078 		 "driver parameters\n");
5079 	return 0;
5080 
5081 	/*
5082 	 * Something bad happened.  Return the error ...
5083 	 */
5084 bye:
5085 	return ret;
5086 }
5087 
5088 /*
5089  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
5090  */
5091 static int adap_init0(struct adapter *adap)
5092 {
5093 	int ret;
5094 	u32 v, port_vec;
5095 	enum dev_state state;
5096 	u32 params[7], val[7];
5097 	struct fw_caps_config_cmd caps_cmd;
5098 	int reset = 1, j;
5099 
5100 	/*
5101 	 * Contact FW, advertising Master capability (and potentially forcing
5102 	 * ourselves as the Master PF if our module parameter force_init is
5103 	 * set).
5104 	 */
5105 	ret = t4_fw_hello(adap, adap->mbox, adap->fn,
5106 			  force_init ? MASTER_MUST : MASTER_MAY,
5107 			  &state);
5108 	if (ret < 0) {
5109 		dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
5110 			ret);
5111 		return ret;
5112 	}
5113 	if (ret == adap->mbox)
5114 		adap->flags |= MASTER_PF;
5115 	if (force_init && state == DEV_STATE_INIT)
5116 		state = DEV_STATE_UNINIT;
5117 
5118 	/*
5119 	 * If we're the Master PF Driver and the device is uninitialized,
5120 	 * then let's consider upgrading the firmware ...  (We always want
5121 	 * to check the firmware version number in order to A. get it for
5122 	 * later reporting and B. to warn if the currently loaded firmware
5123 	 * is excessively mismatched relative to the driver.)
5124 	 */
5125 	ret = t4_check_fw_version(adap);
5126 
5127 	/* The error code -EFAULT is returned by t4_check_fw_version() if
5128 	 * firmware on adapter < supported firmware. If firmware on adapter
5129 	 * is too old (not supported by driver) and we're the MASTER_PF set
5130 	 * adapter state to DEV_STATE_UNINIT to force firmware upgrade
5131 	 * and reinitialization.
5132 	 */
5133 	if ((adap->flags & MASTER_PF) && ret == -EFAULT)
5134 		state = DEV_STATE_UNINIT;
5135 	if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
5136 		if (ret == -EINVAL || ret == -EFAULT || ret > 0) {
5137 			if (upgrade_fw(adap) >= 0) {
5138 				/*
5139 				 * Note that the chip was reset as part of the
5140 				 * firmware upgrade so we don't reset it again
5141 				 * below and grab the new firmware version.
5142 				 */
5143 				reset = 0;
5144 				ret = t4_check_fw_version(adap);
5145 			} else
5146 				if (ret == -EFAULT) {
5147 					/*
5148 					 * Firmware is old but still might
5149 					 * work if we force reinitialization
5150 					 * of the adapter. Ignoring FW upgrade
5151 					 * failure.
5152 					 */
5153 					dev_warn(adap->pdev_dev,
5154 						 "Ignoring firmware upgrade "
5155 						 "failure, and forcing driver "
5156 						 "to reinitialize the "
5157 						 "adapter.\n");
5158 					ret = 0;
5159 				}
5160 		}
5161 		if (ret < 0)
5162 			return ret;
5163 	}
5164 
5165 	/*
5166 	 * Grab VPD parameters.  This should be done after we establish a
5167 	 * connection to the firmware since some of the VPD parameters
5168 	 * (notably the Core Clock frequency) are retrieved via requests to
5169 	 * the firmware.  On the other hand, we need these fairly early on
5170 	 * so we do this right after getting ahold of the firmware.
5171 	 */
5172 	ret = get_vpd_params(adap, &adap->params.vpd);
5173 	if (ret < 0)
5174 		goto bye;
5175 
5176 	/*
5177 	 * Find out what ports are available to us.  Note that we need to do
5178 	 * this before calling adap_init0_no_config() since it needs nports
5179 	 * and portvec ...
5180 	 */
5181 	v =
5182 	    FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5183 	    FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
5184 	ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
5185 	if (ret < 0)
5186 		goto bye;
5187 
5188 	adap->params.nports = hweight32(port_vec);
5189 	adap->params.portvec = port_vec;
5190 
5191 	/*
5192 	 * If the firmware is initialized already (and we're not forcing a
5193 	 * master initialization), note that we're living with existing
5194 	 * adapter parameters.  Otherwise, it's time to try initializing the
5195 	 * adapter ...
5196 	 */
5197 	if (state == DEV_STATE_INIT) {
5198 		dev_info(adap->pdev_dev, "Coming up as %s: "\
5199 			 "Adapter already initialized\n",
5200 			 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
5201 		adap->flags |= USING_SOFT_PARAMS;
5202 	} else {
5203 		dev_info(adap->pdev_dev, "Coming up as MASTER: "\
5204 			 "Initializing adapter\n");
5205 
5206 		/*
5207 		 * If the firmware doesn't support Configuration
5208 		 * Files warn user and exit,
5209 		 */
5210 		if (ret < 0)
5211 			dev_warn(adap->pdev_dev, "Firmware doesn't support "
5212 				 "configuration file.\n");
5213 		if (force_old_init)
5214 			ret = adap_init0_no_config(adap, reset);
5215 		else {
5216 			/*
5217 			 * Find out whether we're dealing with a version of
5218 			 * the firmware which has configuration file support.
5219 			 */
5220 			params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5221 				     FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
5222 			ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
5223 					      params, val);
5224 
5225 			/*
5226 			 * If the firmware doesn't support Configuration
5227 			 * Files, use the old Driver-based, hard-wired
5228 			 * initialization.  Otherwise, try using the
5229 			 * Configuration File support and fall back to the
5230 			 * Driver-based initialization if there's no
5231 			 * Configuration File found.
5232 			 */
5233 			if (ret < 0)
5234 				ret = adap_init0_no_config(adap, reset);
5235 			else {
5236 				/*
5237 				 * The firmware provides us with a memory
5238 				 * buffer where we can load a Configuration
5239 				 * File from the host if we want to override
5240 				 * the Configuration File in flash.
5241 				 */
5242 
5243 				ret = adap_init0_config(adap, reset);
5244 				if (ret == -ENOENT) {
5245 					dev_info(adap->pdev_dev,
5246 					    "No Configuration File present "
5247 					    "on adapter.  Using hard-wired "
5248 					    "configuration parameters.\n");
5249 					ret = adap_init0_no_config(adap, reset);
5250 				}
5251 			}
5252 		}
5253 		if (ret < 0) {
5254 			dev_err(adap->pdev_dev,
5255 				"could not initialize adapter, error %d\n",
5256 				-ret);
5257 			goto bye;
5258 		}
5259 	}
5260 
5261 	/*
5262 	 * If we're living with non-hard-coded parameters (either from a
5263 	 * Firmware Configuration File or values programmed by a different PF
5264 	 * Driver), give the SGE code a chance to pull in anything that it
5265 	 * needs ...  Note that this must be called after we retrieve our VPD
5266 	 * parameters in order to know how to convert core ticks to seconds.
5267 	 */
5268 	if (adap->flags & USING_SOFT_PARAMS) {
5269 		ret = t4_sge_init(adap);
5270 		if (ret < 0)
5271 			goto bye;
5272 	}
5273 
5274 	if (is_bypass_device(adap->pdev->device))
5275 		adap->params.bypass = 1;
5276 
5277 	/*
5278 	 * Grab some of our basic fundamental operating parameters.
5279 	 */
5280 #define FW_PARAM_DEV(param) \
5281 	(FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
5282 	FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
5283 
5284 #define FW_PARAM_PFVF(param) \
5285 	FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
5286 	FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)|  \
5287 	FW_PARAMS_PARAM_Y(0) | \
5288 	FW_PARAMS_PARAM_Z(0)
5289 
5290 	params[0] = FW_PARAM_PFVF(EQ_START);
5291 	params[1] = FW_PARAM_PFVF(L2T_START);
5292 	params[2] = FW_PARAM_PFVF(L2T_END);
5293 	params[3] = FW_PARAM_PFVF(FILTER_START);
5294 	params[4] = FW_PARAM_PFVF(FILTER_END);
5295 	params[5] = FW_PARAM_PFVF(IQFLINT_START);
5296 	ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
5297 	if (ret < 0)
5298 		goto bye;
5299 	adap->sge.egr_start = val[0];
5300 	adap->l2t_start = val[1];
5301 	adap->l2t_end = val[2];
5302 	adap->tids.ftid_base = val[3];
5303 	adap->tids.nftids = val[4] - val[3] + 1;
5304 	adap->sge.ingr_start = val[5];
5305 
5306 	/* query params related to active filter region */
5307 	params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
5308 	params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
5309 	ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
5310 	/* If Active filter size is set we enable establishing
5311 	 * offload connection through firmware work request
5312 	 */
5313 	if ((val[0] != val[1]) && (ret >= 0)) {
5314 		adap->flags |= FW_OFLD_CONN;
5315 		adap->tids.aftid_base = val[0];
5316 		adap->tids.aftid_end = val[1];
5317 	}
5318 
5319 	/* If we're running on newer firmware, let it know that we're
5320 	 * prepared to deal with encapsulated CPL messages.  Older
5321 	 * firmware won't understand this and we'll just get
5322 	 * unencapsulated messages ...
5323 	 */
5324 	params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
5325 	val[0] = 1;
5326 	(void) t4_set_params(adap, adap->mbox, adap->fn, 0, 1, params, val);
5327 
5328 	/*
5329 	 * Get device capabilities so we can determine what resources we need
5330 	 * to manage.
5331 	 */
5332 	memset(&caps_cmd, 0, sizeof(caps_cmd));
5333 	caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5334 				     FW_CMD_REQUEST | FW_CMD_READ);
5335 	caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5336 	ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
5337 			 &caps_cmd);
5338 	if (ret < 0)
5339 		goto bye;
5340 
5341 	if (caps_cmd.ofldcaps) {
5342 		/* query offload-related parameters */
5343 		params[0] = FW_PARAM_DEV(NTID);
5344 		params[1] = FW_PARAM_PFVF(SERVER_START);
5345 		params[2] = FW_PARAM_PFVF(SERVER_END);
5346 		params[3] = FW_PARAM_PFVF(TDDP_START);
5347 		params[4] = FW_PARAM_PFVF(TDDP_END);
5348 		params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5349 		ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5350 				      params, val);
5351 		if (ret < 0)
5352 			goto bye;
5353 		adap->tids.ntids = val[0];
5354 		adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
5355 		adap->tids.stid_base = val[1];
5356 		adap->tids.nstids = val[2] - val[1] + 1;
5357 		/*
5358 		 * Setup server filter region. Divide the availble filter
5359 		 * region into two parts. Regular filters get 1/3rd and server
5360 		 * filters get 2/3rd part. This is only enabled if workarond
5361 		 * path is enabled.
5362 		 * 1. For regular filters.
5363 		 * 2. Server filter: This are special filters which are used
5364 		 * to redirect SYN packets to offload queue.
5365 		 */
5366 		if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
5367 			adap->tids.sftid_base = adap->tids.ftid_base +
5368 					DIV_ROUND_UP(adap->tids.nftids, 3);
5369 			adap->tids.nsftids = adap->tids.nftids -
5370 					 DIV_ROUND_UP(adap->tids.nftids, 3);
5371 			adap->tids.nftids = adap->tids.sftid_base -
5372 						adap->tids.ftid_base;
5373 		}
5374 		adap->vres.ddp.start = val[3];
5375 		adap->vres.ddp.size = val[4] - val[3] + 1;
5376 		adap->params.ofldq_wr_cred = val[5];
5377 
5378 		adap->params.offload = 1;
5379 	}
5380 	if (caps_cmd.rdmacaps) {
5381 		params[0] = FW_PARAM_PFVF(STAG_START);
5382 		params[1] = FW_PARAM_PFVF(STAG_END);
5383 		params[2] = FW_PARAM_PFVF(RQ_START);
5384 		params[3] = FW_PARAM_PFVF(RQ_END);
5385 		params[4] = FW_PARAM_PFVF(PBL_START);
5386 		params[5] = FW_PARAM_PFVF(PBL_END);
5387 		ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5388 				      params, val);
5389 		if (ret < 0)
5390 			goto bye;
5391 		adap->vres.stag.start = val[0];
5392 		adap->vres.stag.size = val[1] - val[0] + 1;
5393 		adap->vres.rq.start = val[2];
5394 		adap->vres.rq.size = val[3] - val[2] + 1;
5395 		adap->vres.pbl.start = val[4];
5396 		adap->vres.pbl.size = val[5] - val[4] + 1;
5397 
5398 		params[0] = FW_PARAM_PFVF(SQRQ_START);
5399 		params[1] = FW_PARAM_PFVF(SQRQ_END);
5400 		params[2] = FW_PARAM_PFVF(CQ_START);
5401 		params[3] = FW_PARAM_PFVF(CQ_END);
5402 		params[4] = FW_PARAM_PFVF(OCQ_START);
5403 		params[5] = FW_PARAM_PFVF(OCQ_END);
5404 		ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
5405 		if (ret < 0)
5406 			goto bye;
5407 		adap->vres.qp.start = val[0];
5408 		adap->vres.qp.size = val[1] - val[0] + 1;
5409 		adap->vres.cq.start = val[2];
5410 		adap->vres.cq.size = val[3] - val[2] + 1;
5411 		adap->vres.ocq.start = val[4];
5412 		adap->vres.ocq.size = val[5] - val[4] + 1;
5413 	}
5414 	if (caps_cmd.iscsicaps) {
5415 		params[0] = FW_PARAM_PFVF(ISCSI_START);
5416 		params[1] = FW_PARAM_PFVF(ISCSI_END);
5417 		ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
5418 				      params, val);
5419 		if (ret < 0)
5420 			goto bye;
5421 		adap->vres.iscsi.start = val[0];
5422 		adap->vres.iscsi.size = val[1] - val[0] + 1;
5423 	}
5424 #undef FW_PARAM_PFVF
5425 #undef FW_PARAM_DEV
5426 
5427 	/*
5428 	 * These are finalized by FW initialization, load their values now.
5429 	 */
5430 	v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
5431 	adap->params.tp.tre = TIMERRESOLUTION_GET(v);
5432 	adap->params.tp.dack_re = DELAYEDACKRESOLUTION_GET(v);
5433 	t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5434 	t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5435 		     adap->params.b_wnd);
5436 
5437 	/* MODQ_REQ_MAP defaults to setting queues 0-3 to chan 0-3 */
5438 	for (j = 0; j < NCHAN; j++)
5439 		adap->params.tp.tx_modq[j] = j;
5440 
5441 	t4_read_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
5442 			 &adap->filter_mode, 1,
5443 			 TP_VLAN_PRI_MAP);
5444 
5445 	adap->flags |= FW_OK;
5446 	return 0;
5447 
5448 	/*
5449 	 * Something bad happened.  If a command timed out or failed with EIO
5450 	 * FW does not operate within its spec or something catastrophic
5451 	 * happened to HW/FW, stop issuing commands.
5452 	 */
5453 bye:
5454 	if (ret != -ETIMEDOUT && ret != -EIO)
5455 		t4_fw_bye(adap, adap->mbox);
5456 	return ret;
5457 }
5458 
5459 /* EEH callbacks */
5460 
5461 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5462 					 pci_channel_state_t state)
5463 {
5464 	int i;
5465 	struct adapter *adap = pci_get_drvdata(pdev);
5466 
5467 	if (!adap)
5468 		goto out;
5469 
5470 	rtnl_lock();
5471 	adap->flags &= ~FW_OK;
5472 	notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5473 	for_each_port(adap, i) {
5474 		struct net_device *dev = adap->port[i];
5475 
5476 		netif_device_detach(dev);
5477 		netif_carrier_off(dev);
5478 	}
5479 	if (adap->flags & FULL_INIT_DONE)
5480 		cxgb_down(adap);
5481 	rtnl_unlock();
5482 	pci_disable_device(pdev);
5483 out:	return state == pci_channel_io_perm_failure ?
5484 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5485 }
5486 
5487 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5488 {
5489 	int i, ret;
5490 	struct fw_caps_config_cmd c;
5491 	struct adapter *adap = pci_get_drvdata(pdev);
5492 
5493 	if (!adap) {
5494 		pci_restore_state(pdev);
5495 		pci_save_state(pdev);
5496 		return PCI_ERS_RESULT_RECOVERED;
5497 	}
5498 
5499 	if (pci_enable_device(pdev)) {
5500 		dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
5501 		return PCI_ERS_RESULT_DISCONNECT;
5502 	}
5503 
5504 	pci_set_master(pdev);
5505 	pci_restore_state(pdev);
5506 	pci_save_state(pdev);
5507 	pci_cleanup_aer_uncorrect_error_status(pdev);
5508 
5509 	if (t4_wait_dev_ready(adap) < 0)
5510 		return PCI_ERS_RESULT_DISCONNECT;
5511 	if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL) < 0)
5512 		return PCI_ERS_RESULT_DISCONNECT;
5513 	adap->flags |= FW_OK;
5514 	if (adap_init1(adap, &c))
5515 		return PCI_ERS_RESULT_DISCONNECT;
5516 
5517 	for_each_port(adap, i) {
5518 		struct port_info *p = adap2pinfo(adap, i);
5519 
5520 		ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5521 				  NULL, NULL);
5522 		if (ret < 0)
5523 			return PCI_ERS_RESULT_DISCONNECT;
5524 		p->viid = ret;
5525 		p->xact_addr_filt = -1;
5526 	}
5527 
5528 	t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5529 		     adap->params.b_wnd);
5530 	setup_memwin(adap);
5531 	if (cxgb_up(adap))
5532 		return PCI_ERS_RESULT_DISCONNECT;
5533 	return PCI_ERS_RESULT_RECOVERED;
5534 }
5535 
5536 static void eeh_resume(struct pci_dev *pdev)
5537 {
5538 	int i;
5539 	struct adapter *adap = pci_get_drvdata(pdev);
5540 
5541 	if (!adap)
5542 		return;
5543 
5544 	rtnl_lock();
5545 	for_each_port(adap, i) {
5546 		struct net_device *dev = adap->port[i];
5547 
5548 		if (netif_running(dev)) {
5549 			link_start(dev);
5550 			cxgb_set_rxmode(dev);
5551 		}
5552 		netif_device_attach(dev);
5553 	}
5554 	rtnl_unlock();
5555 }
5556 
5557 static const struct pci_error_handlers cxgb4_eeh = {
5558 	.error_detected = eeh_err_detected,
5559 	.slot_reset     = eeh_slot_reset,
5560 	.resume         = eeh_resume,
5561 };
5562 
5563 static inline bool is_10g_port(const struct link_config *lc)
5564 {
5565 	return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
5566 }
5567 
5568 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
5569 			     unsigned int size, unsigned int iqe_size)
5570 {
5571 	q->intr_params = QINTR_TIMER_IDX(timer_idx) |
5572 			 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
5573 	q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
5574 	q->iqe_len = iqe_size;
5575 	q->size = size;
5576 }
5577 
5578 /*
5579  * Perform default configuration of DMA queues depending on the number and type
5580  * of ports we found and the number of available CPUs.  Most settings can be
5581  * modified by the admin prior to actual use.
5582  */
5583 static void cfg_queues(struct adapter *adap)
5584 {
5585 	struct sge *s = &adap->sge;
5586 	int i, q10g = 0, n10g = 0, qidx = 0;
5587 
5588 	for_each_port(adap, i)
5589 		n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
5590 
5591 	/*
5592 	 * We default to 1 queue per non-10G port and up to # of cores queues
5593 	 * per 10G port.
5594 	 */
5595 	if (n10g)
5596 		q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5597 	if (q10g > netif_get_num_default_rss_queues())
5598 		q10g = netif_get_num_default_rss_queues();
5599 
5600 	for_each_port(adap, i) {
5601 		struct port_info *pi = adap2pinfo(adap, i);
5602 
5603 		pi->first_qset = qidx;
5604 		pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
5605 		qidx += pi->nqsets;
5606 	}
5607 
5608 	s->ethqsets = qidx;
5609 	s->max_ethqsets = qidx;   /* MSI-X may lower it later */
5610 
5611 	if (is_offload(adap)) {
5612 		/*
5613 		 * For offload we use 1 queue/channel if all ports are up to 1G,
5614 		 * otherwise we divide all available queues amongst the channels
5615 		 * capped by the number of available cores.
5616 		 */
5617 		if (n10g) {
5618 			i = min_t(int, ARRAY_SIZE(s->ofldrxq),
5619 				  num_online_cpus());
5620 			s->ofldqsets = roundup(i, adap->params.nports);
5621 		} else
5622 			s->ofldqsets = adap->params.nports;
5623 		/* For RDMA one Rx queue per channel suffices */
5624 		s->rdmaqs = adap->params.nports;
5625 	}
5626 
5627 	for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5628 		struct sge_eth_rxq *r = &s->ethrxq[i];
5629 
5630 		init_rspq(&r->rspq, 0, 0, 1024, 64);
5631 		r->fl.size = 72;
5632 	}
5633 
5634 	for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5635 		s->ethtxq[i].q.size = 1024;
5636 
5637 	for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5638 		s->ctrlq[i].q.size = 512;
5639 
5640 	for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
5641 		s->ofldtxq[i].q.size = 1024;
5642 
5643 	for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
5644 		struct sge_ofld_rxq *r = &s->ofldrxq[i];
5645 
5646 		init_rspq(&r->rspq, 0, 0, 1024, 64);
5647 		r->rspq.uld = CXGB4_ULD_ISCSI;
5648 		r->fl.size = 72;
5649 	}
5650 
5651 	for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
5652 		struct sge_ofld_rxq *r = &s->rdmarxq[i];
5653 
5654 		init_rspq(&r->rspq, 0, 0, 511, 64);
5655 		r->rspq.uld = CXGB4_ULD_RDMA;
5656 		r->fl.size = 72;
5657 	}
5658 
5659 	init_rspq(&s->fw_evtq, 6, 0, 512, 64);
5660 	init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
5661 }
5662 
5663 /*
5664  * Reduce the number of Ethernet queues across all ports to at most n.
5665  * n provides at least one queue per port.
5666  */
5667 static void reduce_ethqs(struct adapter *adap, int n)
5668 {
5669 	int i;
5670 	struct port_info *pi;
5671 
5672 	while (n < adap->sge.ethqsets)
5673 		for_each_port(adap, i) {
5674 			pi = adap2pinfo(adap, i);
5675 			if (pi->nqsets > 1) {
5676 				pi->nqsets--;
5677 				adap->sge.ethqsets--;
5678 				if (adap->sge.ethqsets <= n)
5679 					break;
5680 			}
5681 		}
5682 
5683 	n = 0;
5684 	for_each_port(adap, i) {
5685 		pi = adap2pinfo(adap, i);
5686 		pi->first_qset = n;
5687 		n += pi->nqsets;
5688 	}
5689 }
5690 
5691 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5692 #define EXTRA_VECS 2
5693 
5694 static int enable_msix(struct adapter *adap)
5695 {
5696 	int ofld_need = 0;
5697 	int i, err, want, need;
5698 	struct sge *s = &adap->sge;
5699 	unsigned int nchan = adap->params.nports;
5700 	struct msix_entry entries[MAX_INGQ + 1];
5701 
5702 	for (i = 0; i < ARRAY_SIZE(entries); ++i)
5703 		entries[i].entry = i;
5704 
5705 	want = s->max_ethqsets + EXTRA_VECS;
5706 	if (is_offload(adap)) {
5707 		want += s->rdmaqs + s->ofldqsets;
5708 		/* need nchan for each possible ULD */
5709 		ofld_need = 2 * nchan;
5710 	}
5711 	need = adap->params.nports + EXTRA_VECS + ofld_need;
5712 
5713 	while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
5714 		want = err;
5715 
5716 	if (!err) {
5717 		/*
5718 		 * Distribute available vectors to the various queue groups.
5719 		 * Every group gets its minimum requirement and NIC gets top
5720 		 * priority for leftovers.
5721 		 */
5722 		i = want - EXTRA_VECS - ofld_need;
5723 		if (i < s->max_ethqsets) {
5724 			s->max_ethqsets = i;
5725 			if (i < s->ethqsets)
5726 				reduce_ethqs(adap, i);
5727 		}
5728 		if (is_offload(adap)) {
5729 			i = want - EXTRA_VECS - s->max_ethqsets;
5730 			i -= ofld_need - nchan;
5731 			s->ofldqsets = (i / nchan) * nchan;  /* round down */
5732 		}
5733 		for (i = 0; i < want; ++i)
5734 			adap->msix_info[i].vec = entries[i].vector;
5735 	} else if (err > 0)
5736 		dev_info(adap->pdev_dev,
5737 			 "only %d MSI-X vectors left, not using MSI-X\n", err);
5738 	return err;
5739 }
5740 
5741 #undef EXTRA_VECS
5742 
5743 static int init_rss(struct adapter *adap)
5744 {
5745 	unsigned int i, j;
5746 
5747 	for_each_port(adap, i) {
5748 		struct port_info *pi = adap2pinfo(adap, i);
5749 
5750 		pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5751 		if (!pi->rss)
5752 			return -ENOMEM;
5753 		for (j = 0; j < pi->rss_size; j++)
5754 			pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
5755 	}
5756 	return 0;
5757 }
5758 
5759 static void print_port_info(const struct net_device *dev)
5760 {
5761 	static const char *base[] = {
5762 		"R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
5763 		"KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
5764 	};
5765 
5766 	char buf[80];
5767 	char *bufp = buf;
5768 	const char *spd = "";
5769 	const struct port_info *pi = netdev_priv(dev);
5770 	const struct adapter *adap = pi->adapter;
5771 
5772 	if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
5773 		spd = " 2.5 GT/s";
5774 	else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
5775 		spd = " 5 GT/s";
5776 
5777 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5778 		bufp += sprintf(bufp, "100/");
5779 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5780 		bufp += sprintf(bufp, "1000/");
5781 	if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
5782 		bufp += sprintf(bufp, "10G/");
5783 	if (bufp != buf)
5784 		--bufp;
5785 	sprintf(bufp, "BASE-%s", base[pi->port_type]);
5786 
5787 	netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5788 		    adap->params.vpd.id,
5789 		    CHELSIO_CHIP_RELEASE(adap->params.rev), buf,
5790 		    is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
5791 		    (adap->flags & USING_MSIX) ? " MSI-X" :
5792 		    (adap->flags & USING_MSI) ? " MSI" : "");
5793 	netdev_info(dev, "S/N: %s, E/C: %s\n",
5794 		    adap->params.vpd.sn, adap->params.vpd.ec);
5795 }
5796 
5797 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
5798 {
5799 	pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
5800 }
5801 
5802 /*
5803  * Free the following resources:
5804  * - memory used for tables
5805  * - MSI/MSI-X
5806  * - net devices
5807  * - resources FW is holding for us
5808  */
5809 static void free_some_resources(struct adapter *adapter)
5810 {
5811 	unsigned int i;
5812 
5813 	t4_free_mem(adapter->l2t);
5814 	t4_free_mem(adapter->tids.tid_tab);
5815 	disable_msi(adapter);
5816 
5817 	for_each_port(adapter, i)
5818 		if (adapter->port[i]) {
5819 			kfree(adap2pinfo(adapter, i)->rss);
5820 			free_netdev(adapter->port[i]);
5821 		}
5822 	if (adapter->flags & FW_OK)
5823 		t4_fw_bye(adapter, adapter->fn);
5824 }
5825 
5826 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5827 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5828 		   NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5829 #define SEGMENT_SIZE 128
5830 
5831 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5832 {
5833 	int func, i, err, s_qpp, qpp, num_seg;
5834 	struct port_info *pi;
5835 	bool highdma = false;
5836 	struct adapter *adapter = NULL;
5837 
5838 	printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5839 
5840 	err = pci_request_regions(pdev, KBUILD_MODNAME);
5841 	if (err) {
5842 		/* Just info, some other driver may have claimed the device. */
5843 		dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5844 		return err;
5845 	}
5846 
5847 	/* We control everything through one PF */
5848 	func = PCI_FUNC(pdev->devfn);
5849 	if (func != ent->driver_data) {
5850 		pci_save_state(pdev);        /* to restore SR-IOV later */
5851 		goto sriov;
5852 	}
5853 
5854 	err = pci_enable_device(pdev);
5855 	if (err) {
5856 		dev_err(&pdev->dev, "cannot enable PCI device\n");
5857 		goto out_release_regions;
5858 	}
5859 
5860 	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5861 		highdma = true;
5862 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5863 		if (err) {
5864 			dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5865 				"coherent allocations\n");
5866 			goto out_disable_device;
5867 		}
5868 	} else {
5869 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5870 		if (err) {
5871 			dev_err(&pdev->dev, "no usable DMA configuration\n");
5872 			goto out_disable_device;
5873 		}
5874 	}
5875 
5876 	pci_enable_pcie_error_reporting(pdev);
5877 	enable_pcie_relaxed_ordering(pdev);
5878 	pci_set_master(pdev);
5879 	pci_save_state(pdev);
5880 
5881 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5882 	if (!adapter) {
5883 		err = -ENOMEM;
5884 		goto out_disable_device;
5885 	}
5886 
5887 	adapter->regs = pci_ioremap_bar(pdev, 0);
5888 	if (!adapter->regs) {
5889 		dev_err(&pdev->dev, "cannot map device registers\n");
5890 		err = -ENOMEM;
5891 		goto out_free_adapter;
5892 	}
5893 
5894 	adapter->pdev = pdev;
5895 	adapter->pdev_dev = &pdev->dev;
5896 	adapter->mbox = func;
5897 	adapter->fn = func;
5898 	adapter->msg_enable = dflt_msg_enable;
5899 	memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5900 
5901 	spin_lock_init(&adapter->stats_lock);
5902 	spin_lock_init(&adapter->tid_release_lock);
5903 
5904 	INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5905 	INIT_WORK(&adapter->db_full_task, process_db_full);
5906 	INIT_WORK(&adapter->db_drop_task, process_db_drop);
5907 
5908 	err = t4_prep_adapter(adapter);
5909 	if (err)
5910 		goto out_unmap_bar0;
5911 
5912 	if (!is_t4(adapter->chip)) {
5913 		s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
5914 		qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
5915 		      SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
5916 		num_seg = PAGE_SIZE / SEGMENT_SIZE;
5917 
5918 		/* Each segment size is 128B. Write coalescing is enabled only
5919 		 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5920 		 * queue is less no of segments that can be accommodated in
5921 		 * a page size.
5922 		 */
5923 		if (qpp > num_seg) {
5924 			dev_err(&pdev->dev,
5925 				"Incorrect number of egress queues per page\n");
5926 			err = -EINVAL;
5927 			goto out_unmap_bar0;
5928 		}
5929 		adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5930 		pci_resource_len(pdev, 2));
5931 		if (!adapter->bar2) {
5932 			dev_err(&pdev->dev, "cannot map device bar2 region\n");
5933 			err = -ENOMEM;
5934 			goto out_unmap_bar0;
5935 		}
5936 	}
5937 
5938 	setup_memwin(adapter);
5939 	err = adap_init0(adapter);
5940 	setup_memwin_rdma(adapter);
5941 	if (err)
5942 		goto out_unmap_bar;
5943 
5944 	for_each_port(adapter, i) {
5945 		struct net_device *netdev;
5946 
5947 		netdev = alloc_etherdev_mq(sizeof(struct port_info),
5948 					   MAX_ETH_QSETS);
5949 		if (!netdev) {
5950 			err = -ENOMEM;
5951 			goto out_free_dev;
5952 		}
5953 
5954 		SET_NETDEV_DEV(netdev, &pdev->dev);
5955 
5956 		adapter->port[i] = netdev;
5957 		pi = netdev_priv(netdev);
5958 		pi->adapter = adapter;
5959 		pi->xact_addr_filt = -1;
5960 		pi->port_id = i;
5961 		netdev->irq = pdev->irq;
5962 
5963 		netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5964 			NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5965 			NETIF_F_RXCSUM | NETIF_F_RXHASH |
5966 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
5967 		if (highdma)
5968 			netdev->hw_features |= NETIF_F_HIGHDMA;
5969 		netdev->features |= netdev->hw_features;
5970 		netdev->vlan_features = netdev->features & VLAN_FEAT;
5971 
5972 		netdev->priv_flags |= IFF_UNICAST_FLT;
5973 
5974 		netdev->netdev_ops = &cxgb4_netdev_ops;
5975 		SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
5976 	}
5977 
5978 	pci_set_drvdata(pdev, adapter);
5979 
5980 	if (adapter->flags & FW_OK) {
5981 		err = t4_port_init(adapter, func, func, 0);
5982 		if (err)
5983 			goto out_free_dev;
5984 	}
5985 
5986 	/*
5987 	 * Configure queues and allocate tables now, they can be needed as
5988 	 * soon as the first register_netdev completes.
5989 	 */
5990 	cfg_queues(adapter);
5991 
5992 	adapter->l2t = t4_init_l2t();
5993 	if (!adapter->l2t) {
5994 		/* We tolerate a lack of L2T, giving up some functionality */
5995 		dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5996 		adapter->params.offload = 0;
5997 	}
5998 
5999 	if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
6000 		dev_warn(&pdev->dev, "could not allocate TID table, "
6001 			 "continuing\n");
6002 		adapter->params.offload = 0;
6003 	}
6004 
6005 	/* See what interrupts we'll be using */
6006 	if (msi > 1 && enable_msix(adapter) == 0)
6007 		adapter->flags |= USING_MSIX;
6008 	else if (msi > 0 && pci_enable_msi(pdev) == 0)
6009 		adapter->flags |= USING_MSI;
6010 
6011 	err = init_rss(adapter);
6012 	if (err)
6013 		goto out_free_dev;
6014 
6015 	/*
6016 	 * The card is now ready to go.  If any errors occur during device
6017 	 * registration we do not fail the whole card but rather proceed only
6018 	 * with the ports we manage to register successfully.  However we must
6019 	 * register at least one net device.
6020 	 */
6021 	for_each_port(adapter, i) {
6022 		pi = adap2pinfo(adapter, i);
6023 		netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
6024 		netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
6025 
6026 		err = register_netdev(adapter->port[i]);
6027 		if (err)
6028 			break;
6029 		adapter->chan_map[pi->tx_chan] = i;
6030 		print_port_info(adapter->port[i]);
6031 	}
6032 	if (i == 0) {
6033 		dev_err(&pdev->dev, "could not register any net devices\n");
6034 		goto out_free_dev;
6035 	}
6036 	if (err) {
6037 		dev_warn(&pdev->dev, "only %d net devices registered\n", i);
6038 		err = 0;
6039 	}
6040 
6041 	if (cxgb4_debugfs_root) {
6042 		adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
6043 							   cxgb4_debugfs_root);
6044 		setup_debugfs(adapter);
6045 	}
6046 
6047 	/* PCIe EEH recovery on powerpc platforms needs fundamental reset */
6048 	pdev->needs_freset = 1;
6049 
6050 	if (is_offload(adapter))
6051 		attach_ulds(adapter);
6052 
6053 sriov:
6054 #ifdef CONFIG_PCI_IOV
6055 	if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
6056 		if (pci_enable_sriov(pdev, num_vf[func]) == 0)
6057 			dev_info(&pdev->dev,
6058 				 "instantiated %u virtual functions\n",
6059 				 num_vf[func]);
6060 #endif
6061 	return 0;
6062 
6063  out_free_dev:
6064 	free_some_resources(adapter);
6065  out_unmap_bar:
6066 	if (!is_t4(adapter->chip))
6067 		iounmap(adapter->bar2);
6068  out_unmap_bar0:
6069 	iounmap(adapter->regs);
6070  out_free_adapter:
6071 	kfree(adapter);
6072  out_disable_device:
6073 	pci_disable_pcie_error_reporting(pdev);
6074 	pci_disable_device(pdev);
6075  out_release_regions:
6076 	pci_release_regions(pdev);
6077 	pci_set_drvdata(pdev, NULL);
6078 	return err;
6079 }
6080 
6081 static void remove_one(struct pci_dev *pdev)
6082 {
6083 	struct adapter *adapter = pci_get_drvdata(pdev);
6084 
6085 #ifdef CONFIG_PCI_IOV
6086 	pci_disable_sriov(pdev);
6087 
6088 #endif
6089 
6090 	if (adapter) {
6091 		int i;
6092 
6093 		if (is_offload(adapter))
6094 			detach_ulds(adapter);
6095 
6096 		for_each_port(adapter, i)
6097 			if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6098 				unregister_netdev(adapter->port[i]);
6099 
6100 		if (adapter->debugfs_root)
6101 			debugfs_remove_recursive(adapter->debugfs_root);
6102 
6103 		/* If we allocated filters, free up state associated with any
6104 		 * valid filters ...
6105 		 */
6106 		if (adapter->tids.ftid_tab) {
6107 			struct filter_entry *f = &adapter->tids.ftid_tab[0];
6108 			for (i = 0; i < (adapter->tids.nftids +
6109 					adapter->tids.nsftids); i++, f++)
6110 				if (f->valid)
6111 					clear_filter(adapter, f);
6112 		}
6113 
6114 		if (adapter->flags & FULL_INIT_DONE)
6115 			cxgb_down(adapter);
6116 
6117 		free_some_resources(adapter);
6118 		iounmap(adapter->regs);
6119 		if (!is_t4(adapter->chip))
6120 			iounmap(adapter->bar2);
6121 		kfree(adapter);
6122 		pci_disable_pcie_error_reporting(pdev);
6123 		pci_disable_device(pdev);
6124 		pci_release_regions(pdev);
6125 		pci_set_drvdata(pdev, NULL);
6126 	} else
6127 		pci_release_regions(pdev);
6128 }
6129 
6130 static struct pci_driver cxgb4_driver = {
6131 	.name     = KBUILD_MODNAME,
6132 	.id_table = cxgb4_pci_tbl,
6133 	.probe    = init_one,
6134 	.remove   = remove_one,
6135 	.err_handler = &cxgb4_eeh,
6136 };
6137 
6138 static int __init cxgb4_init_module(void)
6139 {
6140 	int ret;
6141 
6142 	workq = create_singlethread_workqueue("cxgb4");
6143 	if (!workq)
6144 		return -ENOMEM;
6145 
6146 	/* Debugfs support is optional, just warn if this fails */
6147 	cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6148 	if (!cxgb4_debugfs_root)
6149 		pr_warn("could not create debugfs entry, continuing\n");
6150 
6151 	ret = pci_register_driver(&cxgb4_driver);
6152 	if (ret < 0) {
6153 		debugfs_remove(cxgb4_debugfs_root);
6154 		destroy_workqueue(workq);
6155 	}
6156 
6157 	register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6158 
6159 	return ret;
6160 }
6161 
6162 static void __exit cxgb4_cleanup_module(void)
6163 {
6164 	unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6165 	pci_unregister_driver(&cxgb4_driver);
6166 	debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
6167 	flush_workqueue(workq);
6168 	destroy_workqueue(workq);
6169 }
6170 
6171 module_init(cxgb4_init_module);
6172 module_exit(cxgb4_cleanup_module);
6173