cxgb4: Update driver version and description
[firefly-linux-kernel-4.4.55.git] / drivers / net / ethernet / chelsio / cxgb4 / cxgb4_main.c
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 <asm/uaccess.h>
64
65 #include "cxgb4.h"
66 #include "t4_regs.h"
67 #include "t4_msg.h"
68 #include "t4fw_api.h"
69 #include "l2t.h"
70
71 #define DRV_VERSION "2.0.0-ko"
72 #define DRV_DESC "Chelsio T4/T5 Network Driver"
73
74 /*
75  * Max interrupt hold-off timer value in us.  Queues fall back to this value
76  * under extreme memory pressure so it's largish to give the system time to
77  * recover.
78  */
79 #define MAX_SGE_TIMERVAL 200U
80
81 enum {
82         /*
83          * Physical Function provisioning constants.
84          */
85         PFRES_NVI = 4,                  /* # of Virtual Interfaces */
86         PFRES_NETHCTRL = 128,           /* # of EQs used for ETH or CTRL Qs */
87         PFRES_NIQFLINT = 128,           /* # of ingress Qs/w Free List(s)/intr
88                                          */
89         PFRES_NEQ = 256,                /* # of egress queues */
90         PFRES_NIQ = 0,                  /* # of ingress queues */
91         PFRES_TC = 0,                   /* PCI-E traffic class */
92         PFRES_NEXACTF = 128,            /* # of exact MPS filters */
93
94         PFRES_R_CAPS = FW_CMD_CAP_PF,
95         PFRES_WX_CAPS = FW_CMD_CAP_PF,
96
97 #ifdef CONFIG_PCI_IOV
98         /*
99          * Virtual Function provisioning constants.  We need two extra Ingress
100          * Queues with Interrupt capability to serve as the VF's Firmware
101          * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
102          * neither will have Free Lists associated with them).  For each
103          * Ethernet/Control Egress Queue and for each Free List, we need an
104          * Egress Context.
105          */
106         VFRES_NPORTS = 1,               /* # of "ports" per VF */
107         VFRES_NQSETS = 2,               /* # of "Queue Sets" per VF */
108
109         VFRES_NVI = VFRES_NPORTS,       /* # of Virtual Interfaces */
110         VFRES_NETHCTRL = VFRES_NQSETS,  /* # of EQs used for ETH or CTRL Qs */
111         VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
112         VFRES_NEQ = VFRES_NQSETS*2,     /* # of egress queues */
113         VFRES_NIQ = 0,                  /* # of non-fl/int ingress queues */
114         VFRES_TC = 0,                   /* PCI-E traffic class */
115         VFRES_NEXACTF = 16,             /* # of exact MPS filters */
116
117         VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
118         VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
119 #endif
120 };
121
122 /*
123  * Provide a Port Access Rights Mask for the specified PF/VF.  This is very
124  * static and likely not to be useful in the long run.  We really need to
125  * implement some form of persistent configuration which the firmware
126  * controls.
127  */
128 static unsigned int pfvfres_pmask(struct adapter *adapter,
129                                   unsigned int pf, unsigned int vf)
130 {
131         unsigned int portn, portvec;
132
133         /*
134          * Give PF's access to all of the ports.
135          */
136         if (vf == 0)
137                 return FW_PFVF_CMD_PMASK_MASK;
138
139         /*
140          * For VFs, we'll assign them access to the ports based purely on the
141          * PF.  We assign active ports in order, wrapping around if there are
142          * fewer active ports than PFs: e.g. active port[pf % nports].
143          * Unfortunately the adapter's port_info structs haven't been
144          * initialized yet so we have to compute this.
145          */
146         if (adapter->params.nports == 0)
147                 return 0;
148
149         portn = pf % adapter->params.nports;
150         portvec = adapter->params.portvec;
151         for (;;) {
152                 /*
153                  * Isolate the lowest set bit in the port vector.  If we're at
154                  * the port number that we want, return that as the pmask.
155                  * otherwise mask that bit out of the port vector and
156                  * decrement our port number ...
157                  */
158                 unsigned int pmask = portvec ^ (portvec & (portvec-1));
159                 if (portn == 0)
160                         return pmask;
161                 portn--;
162                 portvec &= ~pmask;
163         }
164         /*NOTREACHED*/
165 }
166
167 enum {
168         MAX_TXQ_ENTRIES      = 16384,
169         MAX_CTRL_TXQ_ENTRIES = 1024,
170         MAX_RSPQ_ENTRIES     = 16384,
171         MAX_RX_BUFFERS       = 16384,
172         MIN_TXQ_ENTRIES      = 32,
173         MIN_CTRL_TXQ_ENTRIES = 32,
174         MIN_RSPQ_ENTRIES     = 128,
175         MIN_FL_ENTRIES       = 16
176 };
177
178 /* Host shadow copy of ingress filter entry.  This is in host native format
179  * and doesn't match the ordering or bit order, etc. of the hardware of the
180  * firmware command.  The use of bit-field structure elements is purely to
181  * remind ourselves of the field size limitations and save memory in the case
182  * where the filter table is large.
183  */
184 struct filter_entry {
185         /* Administrative fields for filter.
186          */
187         u32 valid:1;            /* filter allocated and valid */
188         u32 locked:1;           /* filter is administratively locked */
189
190         u32 pending:1;          /* filter action is pending firmware reply */
191         u32 smtidx:8;           /* Source MAC Table index for smac */
192         struct l2t_entry *l2t;  /* Layer Two Table entry for dmac */
193
194         /* The filter itself.  Most of this is a straight copy of information
195          * provided by the extended ioctl().  Some fields are translated to
196          * internal forms -- for instance the Ingress Queue ID passed in from
197          * the ioctl() is translated into the Absolute Ingress Queue ID.
198          */
199         struct ch_filter_specification fs;
200 };
201
202 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
203                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
204                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
205
206 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
207
208 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
209         CH_DEVICE(0xa000, 0),  /* PE10K */
210         CH_DEVICE(0x4001, -1),
211         CH_DEVICE(0x4002, -1),
212         CH_DEVICE(0x4003, -1),
213         CH_DEVICE(0x4004, -1),
214         CH_DEVICE(0x4005, -1),
215         CH_DEVICE(0x4006, -1),
216         CH_DEVICE(0x4007, -1),
217         CH_DEVICE(0x4008, -1),
218         CH_DEVICE(0x4009, -1),
219         CH_DEVICE(0x400a, -1),
220         CH_DEVICE(0x4401, 4),
221         CH_DEVICE(0x4402, 4),
222         CH_DEVICE(0x4403, 4),
223         CH_DEVICE(0x4404, 4),
224         CH_DEVICE(0x4405, 4),
225         CH_DEVICE(0x4406, 4),
226         CH_DEVICE(0x4407, 4),
227         CH_DEVICE(0x4408, 4),
228         CH_DEVICE(0x4409, 4),
229         CH_DEVICE(0x440a, 4),
230         CH_DEVICE(0x440d, 4),
231         CH_DEVICE(0x440e, 4),
232         CH_DEVICE(0x5001, 5),
233         CH_DEVICE(0x5002, 5),
234         CH_DEVICE(0x5003, 5),
235         CH_DEVICE(0x5004, 5),
236         CH_DEVICE(0x5005, 5),
237         CH_DEVICE(0x5006, 5),
238         CH_DEVICE(0x5007, 5),
239         CH_DEVICE(0x5008, 5),
240         CH_DEVICE(0x5009, 5),
241         CH_DEVICE(0x500A, 5),
242         CH_DEVICE(0x500B, 5),
243         CH_DEVICE(0x500C, 5),
244         CH_DEVICE(0x500D, 5),
245         CH_DEVICE(0x500E, 5),
246         CH_DEVICE(0x500F, 5),
247         CH_DEVICE(0x5010, 5),
248         CH_DEVICE(0x5011, 5),
249         CH_DEVICE(0x5012, 5),
250         CH_DEVICE(0x5013, 5),
251         CH_DEVICE(0x5401, 5),
252         CH_DEVICE(0x5402, 5),
253         CH_DEVICE(0x5403, 5),
254         CH_DEVICE(0x5404, 5),
255         CH_DEVICE(0x5405, 5),
256         CH_DEVICE(0x5406, 5),
257         CH_DEVICE(0x5407, 5),
258         CH_DEVICE(0x5408, 5),
259         CH_DEVICE(0x5409, 5),
260         CH_DEVICE(0x540A, 5),
261         CH_DEVICE(0x540B, 5),
262         CH_DEVICE(0x540C, 5),
263         CH_DEVICE(0x540D, 5),
264         CH_DEVICE(0x540E, 5),
265         CH_DEVICE(0x540F, 5),
266         CH_DEVICE(0x5410, 5),
267         CH_DEVICE(0x5411, 5),
268         CH_DEVICE(0x5412, 5),
269         CH_DEVICE(0x5413, 5),
270         { 0, }
271 };
272
273 #define FW_FNAME "cxgb4/t4fw.bin"
274 #define FW5_FNAME "cxgb4/t5fw.bin"
275 #define FW_CFNAME "cxgb4/t4-config.txt"
276 #define FW5_CFNAME "cxgb4/t5-config.txt"
277
278 MODULE_DESCRIPTION(DRV_DESC);
279 MODULE_AUTHOR("Chelsio Communications");
280 MODULE_LICENSE("Dual BSD/GPL");
281 MODULE_VERSION(DRV_VERSION);
282 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
283 MODULE_FIRMWARE(FW_FNAME);
284 MODULE_FIRMWARE(FW5_FNAME);
285
286 /*
287  * Normally we're willing to become the firmware's Master PF but will be happy
288  * if another PF has already become the Master and initialized the adapter.
289  * Setting "force_init" will cause this driver to forcibly establish itself as
290  * the Master PF and initialize the adapter.
291  */
292 static uint force_init;
293
294 module_param(force_init, uint, 0644);
295 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
296
297 /*
298  * Normally if the firmware we connect to has Configuration File support, we
299  * use that and only fall back to the old Driver-based initialization if the
300  * Configuration File fails for some reason.  If force_old_init is set, then
301  * we'll always use the old Driver-based initialization sequence.
302  */
303 static uint force_old_init;
304
305 module_param(force_old_init, uint, 0644);
306 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
307
308 static int dflt_msg_enable = DFLT_MSG_ENABLE;
309
310 module_param(dflt_msg_enable, int, 0644);
311 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
312
313 /*
314  * The driver uses the best interrupt scheme available on a platform in the
315  * order MSI-X, MSI, legacy INTx interrupts.  This parameter determines which
316  * of these schemes the driver may consider as follows:
317  *
318  * msi = 2: choose from among all three options
319  * msi = 1: only consider MSI and INTx interrupts
320  * msi = 0: force INTx interrupts
321  */
322 static int msi = 2;
323
324 module_param(msi, int, 0644);
325 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
326
327 /*
328  * Queue interrupt hold-off timer values.  Queues default to the first of these
329  * upon creation.
330  */
331 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
332
333 module_param_array(intr_holdoff, uint, NULL, 0644);
334 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
335                  "0..4 in microseconds");
336
337 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
338
339 module_param_array(intr_cnt, uint, NULL, 0644);
340 MODULE_PARM_DESC(intr_cnt,
341                  "thresholds 1..3 for queue interrupt packet counters");
342
343 /*
344  * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
345  * offset by 2 bytes in order to have the IP headers line up on 4-byte
346  * boundaries.  This is a requirement for many architectures which will throw
347  * a machine check fault if an attempt is made to access one of the 4-byte IP
348  * header fields on a non-4-byte boundary.  And it's a major performance issue
349  * even on some architectures which allow it like some implementations of the
350  * x86 ISA.  However, some architectures don't mind this and for some very
351  * edge-case performance sensitive applications (like forwarding large volumes
352  * of small packets), setting this DMA offset to 0 will decrease the number of
353  * PCI-E Bus transfers enough to measurably affect performance.
354  */
355 static int rx_dma_offset = 2;
356
357 static bool vf_acls;
358
359 #ifdef CONFIG_PCI_IOV
360 module_param(vf_acls, bool, 0644);
361 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
362
363 /* Since T5 has more num of PFs, using NUM_OF_PF_WITH_SRIOV_T5
364  * macro as num_vf array size
365  */
366 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV_T5];
367
368 module_param_array(num_vf, uint, NULL, 0644);
369 MODULE_PARM_DESC(num_vf,
370                  "number of VFs for each of PFs 0-3 for T4 and PFs 0-7 for T5");
371 #endif
372
373 /*
374  * The filter TCAM has a fixed portion and a variable portion.  The fixed
375  * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
376  * ports.  The variable portion is 36 bits which can include things like Exact
377  * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
378  * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
379  * far exceed the 36-bit budget for this "compressed" header portion of the
380  * filter.  Thus, we have a scarce resource which must be carefully managed.
381  *
382  * By default we set this up to mostly match the set of filter matching
383  * capabilities of T3 but with accommodations for some of T4's more
384  * interesting features:
385  *
386  *   { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
387  *     [Inner] VLAN (17), Port (3), FCoE (1) }
388  */
389 enum {
390         TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
391         TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
392         TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
393 };
394
395 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
396
397 module_param(tp_vlan_pri_map, uint, 0644);
398 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
399
400 static struct dentry *cxgb4_debugfs_root;
401
402 static LIST_HEAD(adapter_list);
403 static DEFINE_MUTEX(uld_mutex);
404 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
405 static const char *uld_str[] = { "RDMA", "iSCSI" };
406
407 static void link_report(struct net_device *dev)
408 {
409         if (!netif_carrier_ok(dev))
410                 netdev_info(dev, "link down\n");
411         else {
412                 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
413
414                 const char *s = "10Mbps";
415                 const struct port_info *p = netdev_priv(dev);
416
417                 switch (p->link_cfg.speed) {
418                 case SPEED_10000:
419                         s = "10Gbps";
420                         break;
421                 case SPEED_1000:
422                         s = "1000Mbps";
423                         break;
424                 case SPEED_100:
425                         s = "100Mbps";
426                         break;
427                 }
428
429                 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
430                             fc[p->link_cfg.fc]);
431         }
432 }
433
434 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
435 {
436         struct net_device *dev = adapter->port[port_id];
437
438         /* Skip changes from disabled ports. */
439         if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
440                 if (link_stat)
441                         netif_carrier_on(dev);
442                 else
443                         netif_carrier_off(dev);
444
445                 link_report(dev);
446         }
447 }
448
449 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
450 {
451         static const char *mod_str[] = {
452                 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
453         };
454
455         const struct net_device *dev = adap->port[port_id];
456         const struct port_info *pi = netdev_priv(dev);
457
458         if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
459                 netdev_info(dev, "port module unplugged\n");
460         else if (pi->mod_type < ARRAY_SIZE(mod_str))
461                 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
462 }
463
464 /*
465  * Configure the exact and hash address filters to handle a port's multicast
466  * and secondary unicast MAC addresses.
467  */
468 static int set_addr_filters(const struct net_device *dev, bool sleep)
469 {
470         u64 mhash = 0;
471         u64 uhash = 0;
472         bool free = true;
473         u16 filt_idx[7];
474         const u8 *addr[7];
475         int ret, naddr = 0;
476         const struct netdev_hw_addr *ha;
477         int uc_cnt = netdev_uc_count(dev);
478         int mc_cnt = netdev_mc_count(dev);
479         const struct port_info *pi = netdev_priv(dev);
480         unsigned int mb = pi->adapter->fn;
481
482         /* first do the secondary unicast addresses */
483         netdev_for_each_uc_addr(ha, dev) {
484                 addr[naddr++] = ha->addr;
485                 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
486                         ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
487                                         naddr, addr, filt_idx, &uhash, sleep);
488                         if (ret < 0)
489                                 return ret;
490
491                         free = false;
492                         naddr = 0;
493                 }
494         }
495
496         /* next set up the multicast addresses */
497         netdev_for_each_mc_addr(ha, dev) {
498                 addr[naddr++] = ha->addr;
499                 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
500                         ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
501                                         naddr, addr, filt_idx, &mhash, sleep);
502                         if (ret < 0)
503                                 return ret;
504
505                         free = false;
506                         naddr = 0;
507                 }
508         }
509
510         return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
511                                 uhash | mhash, sleep);
512 }
513
514 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
515 module_param(dbfifo_int_thresh, int, 0644);
516 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
517
518 /*
519  * usecs to sleep while draining the dbfifo
520  */
521 static int dbfifo_drain_delay = 1000;
522 module_param(dbfifo_drain_delay, int, 0644);
523 MODULE_PARM_DESC(dbfifo_drain_delay,
524                  "usecs to sleep while draining the dbfifo");
525
526 /*
527  * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
528  * If @mtu is -1 it is left unchanged.
529  */
530 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
531 {
532         int ret;
533         struct port_info *pi = netdev_priv(dev);
534
535         ret = set_addr_filters(dev, sleep_ok);
536         if (ret == 0)
537                 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
538                                     (dev->flags & IFF_PROMISC) ? 1 : 0,
539                                     (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
540                                     sleep_ok);
541         return ret;
542 }
543
544 static struct workqueue_struct *workq;
545
546 /**
547  *      link_start - enable a port
548  *      @dev: the port to enable
549  *
550  *      Performs the MAC and PHY actions needed to enable a port.
551  */
552 static int link_start(struct net_device *dev)
553 {
554         int ret;
555         struct port_info *pi = netdev_priv(dev);
556         unsigned int mb = pi->adapter->fn;
557
558         /*
559          * We do not set address filters and promiscuity here, the stack does
560          * that step explicitly.
561          */
562         ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
563                             !!(dev->features & NETIF_F_HW_VLAN_RX), true);
564         if (ret == 0) {
565                 ret = t4_change_mac(pi->adapter, mb, pi->viid,
566                                     pi->xact_addr_filt, dev->dev_addr, true,
567                                     true);
568                 if (ret >= 0) {
569                         pi->xact_addr_filt = ret;
570                         ret = 0;
571                 }
572         }
573         if (ret == 0)
574                 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
575                                     &pi->link_cfg);
576         if (ret == 0)
577                 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
578         return ret;
579 }
580
581 /* Clear a filter and release any of its resources that we own.  This also
582  * clears the filter's "pending" status.
583  */
584 static void clear_filter(struct adapter *adap, struct filter_entry *f)
585 {
586         /* If the new or old filter have loopback rewriteing rules then we'll
587          * need to free any existing Layer Two Table (L2T) entries of the old
588          * filter rule.  The firmware will handle freeing up any Source MAC
589          * Table (SMT) entries used for rewriting Source MAC Addresses in
590          * loopback rules.
591          */
592         if (f->l2t)
593                 cxgb4_l2t_release(f->l2t);
594
595         /* The zeroing of the filter rule below clears the filter valid,
596          * pending, locked flags, l2t pointer, etc. so it's all we need for
597          * this operation.
598          */
599         memset(f, 0, sizeof(*f));
600 }
601
602 /* Handle a filter write/deletion reply.
603  */
604 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
605 {
606         unsigned int idx = GET_TID(rpl);
607         unsigned int nidx = idx - adap->tids.ftid_base;
608         unsigned int ret;
609         struct filter_entry *f;
610
611         if (idx >= adap->tids.ftid_base && nidx <
612            (adap->tids.nftids + adap->tids.nsftids)) {
613                 idx = nidx;
614                 ret = GET_TCB_COOKIE(rpl->cookie);
615                 f = &adap->tids.ftid_tab[idx];
616
617                 if (ret == FW_FILTER_WR_FLT_DELETED) {
618                         /* Clear the filter when we get confirmation from the
619                          * hardware that the filter has been deleted.
620                          */
621                         clear_filter(adap, f);
622                 } else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
623                         dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
624                                 idx);
625                         clear_filter(adap, f);
626                 } else if (ret == FW_FILTER_WR_FLT_ADDED) {
627                         f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
628                         f->pending = 0;  /* asynchronous setup completed */
629                         f->valid = 1;
630                 } else {
631                         /* Something went wrong.  Issue a warning about the
632                          * problem and clear everything out.
633                          */
634                         dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
635                                 idx, ret);
636                         clear_filter(adap, f);
637                 }
638         }
639 }
640
641 /* Response queue handler for the FW event queue.
642  */
643 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
644                           const struct pkt_gl *gl)
645 {
646         u8 opcode = ((const struct rss_header *)rsp)->opcode;
647
648         rsp++;                                          /* skip RSS header */
649         if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
650                 const struct cpl_sge_egr_update *p = (void *)rsp;
651                 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
652                 struct sge_txq *txq;
653
654                 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
655                 txq->restarts++;
656                 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
657                         struct sge_eth_txq *eq;
658
659                         eq = container_of(txq, struct sge_eth_txq, q);
660                         netif_tx_wake_queue(eq->txq);
661                 } else {
662                         struct sge_ofld_txq *oq;
663
664                         oq = container_of(txq, struct sge_ofld_txq, q);
665                         tasklet_schedule(&oq->qresume_tsk);
666                 }
667         } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
668                 const struct cpl_fw6_msg *p = (void *)rsp;
669
670                 if (p->type == 0)
671                         t4_handle_fw_rpl(q->adap, p->data);
672         } else if (opcode == CPL_L2T_WRITE_RPL) {
673                 const struct cpl_l2t_write_rpl *p = (void *)rsp;
674
675                 do_l2t_write_rpl(q->adap, p);
676         } else if (opcode == CPL_SET_TCB_RPL) {
677                 const struct cpl_set_tcb_rpl *p = (void *)rsp;
678
679                 filter_rpl(q->adap, p);
680         } else
681                 dev_err(q->adap->pdev_dev,
682                         "unexpected CPL %#x on FW event queue\n", opcode);
683         return 0;
684 }
685
686 /**
687  *      uldrx_handler - response queue handler for ULD queues
688  *      @q: the response queue that received the packet
689  *      @rsp: the response queue descriptor holding the offload message
690  *      @gl: the gather list of packet fragments
691  *
692  *      Deliver an ingress offload packet to a ULD.  All processing is done by
693  *      the ULD, we just maintain statistics.
694  */
695 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
696                          const struct pkt_gl *gl)
697 {
698         struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
699
700         if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
701                 rxq->stats.nomem++;
702                 return -1;
703         }
704         if (gl == NULL)
705                 rxq->stats.imm++;
706         else if (gl == CXGB4_MSG_AN)
707                 rxq->stats.an++;
708         else
709                 rxq->stats.pkts++;
710         return 0;
711 }
712
713 static void disable_msi(struct adapter *adapter)
714 {
715         if (adapter->flags & USING_MSIX) {
716                 pci_disable_msix(adapter->pdev);
717                 adapter->flags &= ~USING_MSIX;
718         } else if (adapter->flags & USING_MSI) {
719                 pci_disable_msi(adapter->pdev);
720                 adapter->flags &= ~USING_MSI;
721         }
722 }
723
724 /*
725  * Interrupt handler for non-data events used with MSI-X.
726  */
727 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
728 {
729         struct adapter *adap = cookie;
730
731         u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
732         if (v & PFSW) {
733                 adap->swintr = 1;
734                 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
735         }
736         t4_slow_intr_handler(adap);
737         return IRQ_HANDLED;
738 }
739
740 /*
741  * Name the MSI-X interrupts.
742  */
743 static void name_msix_vecs(struct adapter *adap)
744 {
745         int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
746
747         /* non-data interrupts */
748         snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
749
750         /* FW events */
751         snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
752                  adap->port[0]->name);
753
754         /* Ethernet queues */
755         for_each_port(adap, j) {
756                 struct net_device *d = adap->port[j];
757                 const struct port_info *pi = netdev_priv(d);
758
759                 for (i = 0; i < pi->nqsets; i++, msi_idx++)
760                         snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
761                                  d->name, i);
762         }
763
764         /* offload queues */
765         for_each_ofldrxq(&adap->sge, i)
766                 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
767                          adap->port[0]->name, i);
768
769         for_each_rdmarxq(&adap->sge, i)
770                 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
771                          adap->port[0]->name, i);
772 }
773
774 static int request_msix_queue_irqs(struct adapter *adap)
775 {
776         struct sge *s = &adap->sge;
777         int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
778
779         err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
780                           adap->msix_info[1].desc, &s->fw_evtq);
781         if (err)
782                 return err;
783
784         for_each_ethrxq(s, ethqidx) {
785                 err = request_irq(adap->msix_info[msi_index].vec,
786                                   t4_sge_intr_msix, 0,
787                                   adap->msix_info[msi_index].desc,
788                                   &s->ethrxq[ethqidx].rspq);
789                 if (err)
790                         goto unwind;
791                 msi_index++;
792         }
793         for_each_ofldrxq(s, ofldqidx) {
794                 err = request_irq(adap->msix_info[msi_index].vec,
795                                   t4_sge_intr_msix, 0,
796                                   adap->msix_info[msi_index].desc,
797                                   &s->ofldrxq[ofldqidx].rspq);
798                 if (err)
799                         goto unwind;
800                 msi_index++;
801         }
802         for_each_rdmarxq(s, rdmaqidx) {
803                 err = request_irq(adap->msix_info[msi_index].vec,
804                                   t4_sge_intr_msix, 0,
805                                   adap->msix_info[msi_index].desc,
806                                   &s->rdmarxq[rdmaqidx].rspq);
807                 if (err)
808                         goto unwind;
809                 msi_index++;
810         }
811         return 0;
812
813 unwind:
814         while (--rdmaqidx >= 0)
815                 free_irq(adap->msix_info[--msi_index].vec,
816                          &s->rdmarxq[rdmaqidx].rspq);
817         while (--ofldqidx >= 0)
818                 free_irq(adap->msix_info[--msi_index].vec,
819                          &s->ofldrxq[ofldqidx].rspq);
820         while (--ethqidx >= 0)
821                 free_irq(adap->msix_info[--msi_index].vec,
822                          &s->ethrxq[ethqidx].rspq);
823         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
824         return err;
825 }
826
827 static void free_msix_queue_irqs(struct adapter *adap)
828 {
829         int i, msi_index = 2;
830         struct sge *s = &adap->sge;
831
832         free_irq(adap->msix_info[1].vec, &s->fw_evtq);
833         for_each_ethrxq(s, i)
834                 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
835         for_each_ofldrxq(s, i)
836                 free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
837         for_each_rdmarxq(s, i)
838                 free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
839 }
840
841 /**
842  *      write_rss - write the RSS table for a given port
843  *      @pi: the port
844  *      @queues: array of queue indices for RSS
845  *
846  *      Sets up the portion of the HW RSS table for the port's VI to distribute
847  *      packets to the Rx queues in @queues.
848  */
849 static int write_rss(const struct port_info *pi, const u16 *queues)
850 {
851         u16 *rss;
852         int i, err;
853         const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
854
855         rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
856         if (!rss)
857                 return -ENOMEM;
858
859         /* map the queue indices to queue ids */
860         for (i = 0; i < pi->rss_size; i++, queues++)
861                 rss[i] = q[*queues].rspq.abs_id;
862
863         err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
864                                   pi->rss_size, rss, pi->rss_size);
865         kfree(rss);
866         return err;
867 }
868
869 /**
870  *      setup_rss - configure RSS
871  *      @adap: the adapter
872  *
873  *      Sets up RSS for each port.
874  */
875 static int setup_rss(struct adapter *adap)
876 {
877         int i, err;
878
879         for_each_port(adap, i) {
880                 const struct port_info *pi = adap2pinfo(adap, i);
881
882                 err = write_rss(pi, pi->rss);
883                 if (err)
884                         return err;
885         }
886         return 0;
887 }
888
889 /*
890  * Return the channel of the ingress queue with the given qid.
891  */
892 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
893 {
894         qid -= p->ingr_start;
895         return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
896 }
897
898 /*
899  * Wait until all NAPI handlers are descheduled.
900  */
901 static void quiesce_rx(struct adapter *adap)
902 {
903         int i;
904
905         for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
906                 struct sge_rspq *q = adap->sge.ingr_map[i];
907
908                 if (q && q->handler)
909                         napi_disable(&q->napi);
910         }
911 }
912
913 /*
914  * Enable NAPI scheduling and interrupt generation for all Rx queues.
915  */
916 static void enable_rx(struct adapter *adap)
917 {
918         int i;
919
920         for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
921                 struct sge_rspq *q = adap->sge.ingr_map[i];
922
923                 if (!q)
924                         continue;
925                 if (q->handler)
926                         napi_enable(&q->napi);
927                 /* 0-increment GTS to start the timer and enable interrupts */
928                 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
929                              SEINTARM(q->intr_params) |
930                              INGRESSQID(q->cntxt_id));
931         }
932 }
933
934 /**
935  *      setup_sge_queues - configure SGE Tx/Rx/response queues
936  *      @adap: the adapter
937  *
938  *      Determines how many sets of SGE queues to use and initializes them.
939  *      We support multiple queue sets per port if we have MSI-X, otherwise
940  *      just one queue set per port.
941  */
942 static int setup_sge_queues(struct adapter *adap)
943 {
944         int err, msi_idx, i, j;
945         struct sge *s = &adap->sge;
946
947         bitmap_zero(s->starving_fl, MAX_EGRQ);
948         bitmap_zero(s->txq_maperr, MAX_EGRQ);
949
950         if (adap->flags & USING_MSIX)
951                 msi_idx = 1;         /* vector 0 is for non-queue interrupts */
952         else {
953                 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
954                                        NULL, NULL);
955                 if (err)
956                         return err;
957                 msi_idx = -((int)s->intrq.abs_id + 1);
958         }
959
960         err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
961                                msi_idx, NULL, fwevtq_handler);
962         if (err) {
963 freeout:        t4_free_sge_resources(adap);
964                 return err;
965         }
966
967         for_each_port(adap, i) {
968                 struct net_device *dev = adap->port[i];
969                 struct port_info *pi = netdev_priv(dev);
970                 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
971                 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
972
973                 for (j = 0; j < pi->nqsets; j++, q++) {
974                         if (msi_idx > 0)
975                                 msi_idx++;
976                         err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
977                                                msi_idx, &q->fl,
978                                                t4_ethrx_handler);
979                         if (err)
980                                 goto freeout;
981                         q->rspq.idx = j;
982                         memset(&q->stats, 0, sizeof(q->stats));
983                 }
984                 for (j = 0; j < pi->nqsets; j++, t++) {
985                         err = t4_sge_alloc_eth_txq(adap, t, dev,
986                                         netdev_get_tx_queue(dev, j),
987                                         s->fw_evtq.cntxt_id);
988                         if (err)
989                                 goto freeout;
990                 }
991         }
992
993         j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
994         for_each_ofldrxq(s, i) {
995                 struct sge_ofld_rxq *q = &s->ofldrxq[i];
996                 struct net_device *dev = adap->port[i / j];
997
998                 if (msi_idx > 0)
999                         msi_idx++;
1000                 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
1001                                        &q->fl, uldrx_handler);
1002                 if (err)
1003                         goto freeout;
1004                 memset(&q->stats, 0, sizeof(q->stats));
1005                 s->ofld_rxq[i] = q->rspq.abs_id;
1006                 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
1007                                             s->fw_evtq.cntxt_id);
1008                 if (err)
1009                         goto freeout;
1010         }
1011
1012         for_each_rdmarxq(s, i) {
1013                 struct sge_ofld_rxq *q = &s->rdmarxq[i];
1014
1015                 if (msi_idx > 0)
1016                         msi_idx++;
1017                 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
1018                                        msi_idx, &q->fl, uldrx_handler);
1019                 if (err)
1020                         goto freeout;
1021                 memset(&q->stats, 0, sizeof(q->stats));
1022                 s->rdma_rxq[i] = q->rspq.abs_id;
1023         }
1024
1025         for_each_port(adap, i) {
1026                 /*
1027                  * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1028                  * have RDMA queues, and that's the right value.
1029                  */
1030                 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1031                                             s->fw_evtq.cntxt_id,
1032                                             s->rdmarxq[i].rspq.cntxt_id);
1033                 if (err)
1034                         goto freeout;
1035         }
1036
1037         t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1038                      RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1039                      QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1040         return 0;
1041 }
1042
1043 /*
1044  * Returns 0 if new FW was successfully loaded, a positive errno if a load was
1045  * started but failed, and a negative errno if flash load couldn't start.
1046  */
1047 static int upgrade_fw(struct adapter *adap)
1048 {
1049         int ret;
1050         u32 vers, exp_major;
1051         const struct fw_hdr *hdr;
1052         const struct firmware *fw;
1053         struct device *dev = adap->pdev_dev;
1054         char *fw_file_name;
1055
1056         switch (CHELSIO_CHIP_VERSION(adap->chip)) {
1057         case CHELSIO_T4:
1058                 fw_file_name = FW_FNAME;
1059                 exp_major = FW_VERSION_MAJOR;
1060                 break;
1061         case CHELSIO_T5:
1062                 fw_file_name = FW5_FNAME;
1063                 exp_major = FW_VERSION_MAJOR_T5;
1064                 break;
1065         default:
1066                 dev_err(dev, "Unsupported chip type, %x\n", adap->chip);
1067                 return -EINVAL;
1068         }
1069
1070         ret = request_firmware(&fw, fw_file_name, dev);
1071         if (ret < 0) {
1072                 dev_err(dev, "unable to load firmware image %s, error %d\n",
1073                         fw_file_name, ret);
1074                 return ret;
1075         }
1076
1077         hdr = (const struct fw_hdr *)fw->data;
1078         vers = ntohl(hdr->fw_ver);
1079         if (FW_HDR_FW_VER_MAJOR_GET(vers) != exp_major) {
1080                 ret = -EINVAL;              /* wrong major version, won't do */
1081                 goto out;
1082         }
1083
1084         /*
1085          * If the flash FW is unusable or we found something newer, load it.
1086          */
1087         if (FW_HDR_FW_VER_MAJOR_GET(adap->params.fw_vers) != exp_major ||
1088             vers > adap->params.fw_vers) {
1089                 dev_info(dev, "upgrading firmware ...\n");
1090                 ret = t4_fw_upgrade(adap, adap->mbox, fw->data, fw->size,
1091                                     /*force=*/false);
1092                 if (!ret)
1093                         dev_info(dev,
1094                                  "firmware upgraded to version %pI4 from %s\n",
1095                                  &hdr->fw_ver, fw_file_name);
1096                 else
1097                         dev_err(dev, "firmware upgrade failed! err=%d\n", -ret);
1098         } else {
1099                 /*
1100                  * Tell our caller that we didn't upgrade the firmware.
1101                  */
1102                 ret = -EINVAL;
1103         }
1104
1105 out:    release_firmware(fw);
1106         return ret;
1107 }
1108
1109 /*
1110  * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1111  * The allocated memory is cleared.
1112  */
1113 void *t4_alloc_mem(size_t size)
1114 {
1115         void *p = kzalloc(size, GFP_KERNEL);
1116
1117         if (!p)
1118                 p = vzalloc(size);
1119         return p;
1120 }
1121
1122 /*
1123  * Free memory allocated through alloc_mem().
1124  */
1125 static void t4_free_mem(void *addr)
1126 {
1127         if (is_vmalloc_addr(addr))
1128                 vfree(addr);
1129         else
1130                 kfree(addr);
1131 }
1132
1133 /* Send a Work Request to write the filter at a specified index.  We construct
1134  * a Firmware Filter Work Request to have the work done and put the indicated
1135  * filter into "pending" mode which will prevent any further actions against
1136  * it till we get a reply from the firmware on the completion status of the
1137  * request.
1138  */
1139 static int set_filter_wr(struct adapter *adapter, int fidx)
1140 {
1141         struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1142         struct sk_buff *skb;
1143         struct fw_filter_wr *fwr;
1144         unsigned int ftid;
1145
1146         /* If the new filter requires loopback Destination MAC and/or VLAN
1147          * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1148          * the filter.
1149          */
1150         if (f->fs.newdmac || f->fs.newvlan) {
1151                 /* allocate L2T entry for new filter */
1152                 f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1153                 if (f->l2t == NULL)
1154                         return -EAGAIN;
1155                 if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1156                                         f->fs.eport, f->fs.dmac)) {
1157                         cxgb4_l2t_release(f->l2t);
1158                         f->l2t = NULL;
1159                         return -ENOMEM;
1160                 }
1161         }
1162
1163         ftid = adapter->tids.ftid_base + fidx;
1164
1165         skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1166         fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1167         memset(fwr, 0, sizeof(*fwr));
1168
1169         /* It would be nice to put most of the following in t4_hw.c but most
1170          * of the work is translating the cxgbtool ch_filter_specification
1171          * into the Work Request and the definition of that structure is
1172          * currently in cxgbtool.h which isn't appropriate to pull into the
1173          * common code.  We may eventually try to come up with a more neutral
1174          * filter specification structure but for now it's easiest to simply
1175          * put this fairly direct code in line ...
1176          */
1177         fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1178         fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1179         fwr->tid_to_iq =
1180                 htonl(V_FW_FILTER_WR_TID(ftid) |
1181                       V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1182                       V_FW_FILTER_WR_NOREPLY(0) |
1183                       V_FW_FILTER_WR_IQ(f->fs.iq));
1184         fwr->del_filter_to_l2tix =
1185                 htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1186                       V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1187                       V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1188                       V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1189                       V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1190                       V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1191                       V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1192                       V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1193                       V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1194                                              f->fs.newvlan == VLAN_REWRITE) |
1195                       V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1196                                             f->fs.newvlan == VLAN_REWRITE) |
1197                       V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1198                       V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1199                       V_FW_FILTER_WR_PRIO(f->fs.prio) |
1200                       V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1201         fwr->ethtype = htons(f->fs.val.ethtype);
1202         fwr->ethtypem = htons(f->fs.mask.ethtype);
1203         fwr->frag_to_ovlan_vldm =
1204                 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1205                  V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1206                  V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1207                  V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1208                  V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1209                  V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1210         fwr->smac_sel = 0;
1211         fwr->rx_chan_rx_rpl_iq =
1212                 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1213                       V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1214         fwr->maci_to_matchtypem =
1215                 htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1216                       V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1217                       V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1218                       V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1219                       V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1220                       V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1221                       V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1222                       V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1223         fwr->ptcl = f->fs.val.proto;
1224         fwr->ptclm = f->fs.mask.proto;
1225         fwr->ttyp = f->fs.val.tos;
1226         fwr->ttypm = f->fs.mask.tos;
1227         fwr->ivlan = htons(f->fs.val.ivlan);
1228         fwr->ivlanm = htons(f->fs.mask.ivlan);
1229         fwr->ovlan = htons(f->fs.val.ovlan);
1230         fwr->ovlanm = htons(f->fs.mask.ovlan);
1231         memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1232         memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1233         memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1234         memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1235         fwr->lp = htons(f->fs.val.lport);
1236         fwr->lpm = htons(f->fs.mask.lport);
1237         fwr->fp = htons(f->fs.val.fport);
1238         fwr->fpm = htons(f->fs.mask.fport);
1239         if (f->fs.newsmac)
1240                 memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1241
1242         /* Mark the filter as "pending" and ship off the Filter Work Request.
1243          * When we get the Work Request Reply we'll clear the pending status.
1244          */
1245         f->pending = 1;
1246         set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1247         t4_ofld_send(adapter, skb);
1248         return 0;
1249 }
1250
1251 /* Delete the filter at a specified index.
1252  */
1253 static int del_filter_wr(struct adapter *adapter, int fidx)
1254 {
1255         struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1256         struct sk_buff *skb;
1257         struct fw_filter_wr *fwr;
1258         unsigned int len, ftid;
1259
1260         len = sizeof(*fwr);
1261         ftid = adapter->tids.ftid_base + fidx;
1262
1263         skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1264         fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1265         t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1266
1267         /* Mark the filter as "pending" and ship off the Filter Work Request.
1268          * When we get the Work Request Reply we'll clear the pending status.
1269          */
1270         f->pending = 1;
1271         t4_mgmt_tx(adapter, skb);
1272         return 0;
1273 }
1274
1275 static inline int is_offload(const struct adapter *adap)
1276 {
1277         return adap->params.offload;
1278 }
1279
1280 /*
1281  * Implementation of ethtool operations.
1282  */
1283
1284 static u32 get_msglevel(struct net_device *dev)
1285 {
1286         return netdev2adap(dev)->msg_enable;
1287 }
1288
1289 static void set_msglevel(struct net_device *dev, u32 val)
1290 {
1291         netdev2adap(dev)->msg_enable = val;
1292 }
1293
1294 static char stats_strings[][ETH_GSTRING_LEN] = {
1295         "TxOctetsOK         ",
1296         "TxFramesOK         ",
1297         "TxBroadcastFrames  ",
1298         "TxMulticastFrames  ",
1299         "TxUnicastFrames    ",
1300         "TxErrorFrames      ",
1301
1302         "TxFrames64         ",
1303         "TxFrames65To127    ",
1304         "TxFrames128To255   ",
1305         "TxFrames256To511   ",
1306         "TxFrames512To1023  ",
1307         "TxFrames1024To1518 ",
1308         "TxFrames1519ToMax  ",
1309
1310         "TxFramesDropped    ",
1311         "TxPauseFrames      ",
1312         "TxPPP0Frames       ",
1313         "TxPPP1Frames       ",
1314         "TxPPP2Frames       ",
1315         "TxPPP3Frames       ",
1316         "TxPPP4Frames       ",
1317         "TxPPP5Frames       ",
1318         "TxPPP6Frames       ",
1319         "TxPPP7Frames       ",
1320
1321         "RxOctetsOK         ",
1322         "RxFramesOK         ",
1323         "RxBroadcastFrames  ",
1324         "RxMulticastFrames  ",
1325         "RxUnicastFrames    ",
1326
1327         "RxFramesTooLong    ",
1328         "RxJabberErrors     ",
1329         "RxFCSErrors        ",
1330         "RxLengthErrors     ",
1331         "RxSymbolErrors     ",
1332         "RxRuntFrames       ",
1333
1334         "RxFrames64         ",
1335         "RxFrames65To127    ",
1336         "RxFrames128To255   ",
1337         "RxFrames256To511   ",
1338         "RxFrames512To1023  ",
1339         "RxFrames1024To1518 ",
1340         "RxFrames1519ToMax  ",
1341
1342         "RxPauseFrames      ",
1343         "RxPPP0Frames       ",
1344         "RxPPP1Frames       ",
1345         "RxPPP2Frames       ",
1346         "RxPPP3Frames       ",
1347         "RxPPP4Frames       ",
1348         "RxPPP5Frames       ",
1349         "RxPPP6Frames       ",
1350         "RxPPP7Frames       ",
1351
1352         "RxBG0FramesDropped ",
1353         "RxBG1FramesDropped ",
1354         "RxBG2FramesDropped ",
1355         "RxBG3FramesDropped ",
1356         "RxBG0FramesTrunc   ",
1357         "RxBG1FramesTrunc   ",
1358         "RxBG2FramesTrunc   ",
1359         "RxBG3FramesTrunc   ",
1360
1361         "TSO                ",
1362         "TxCsumOffload      ",
1363         "RxCsumGood         ",
1364         "VLANextractions    ",
1365         "VLANinsertions     ",
1366         "GROpackets         ",
1367         "GROmerged          ",
1368         "WriteCoalSuccess   ",
1369         "WriteCoalFail      ",
1370 };
1371
1372 static int get_sset_count(struct net_device *dev, int sset)
1373 {
1374         switch (sset) {
1375         case ETH_SS_STATS:
1376                 return ARRAY_SIZE(stats_strings);
1377         default:
1378                 return -EOPNOTSUPP;
1379         }
1380 }
1381
1382 #define T4_REGMAP_SIZE (160 * 1024)
1383 #define T5_REGMAP_SIZE (332 * 1024)
1384
1385 static int get_regs_len(struct net_device *dev)
1386 {
1387         struct adapter *adap = netdev2adap(dev);
1388         if (is_t4(adap->chip))
1389                 return T4_REGMAP_SIZE;
1390         else
1391                 return T5_REGMAP_SIZE;
1392 }
1393
1394 static int get_eeprom_len(struct net_device *dev)
1395 {
1396         return EEPROMSIZE;
1397 }
1398
1399 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1400 {
1401         struct adapter *adapter = netdev2adap(dev);
1402
1403         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1404         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1405         strlcpy(info->bus_info, pci_name(adapter->pdev),
1406                 sizeof(info->bus_info));
1407
1408         if (adapter->params.fw_vers)
1409                 snprintf(info->fw_version, sizeof(info->fw_version),
1410                         "%u.%u.%u.%u, TP %u.%u.%u.%u",
1411                         FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1412                         FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1413                         FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1414                         FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1415                         FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1416                         FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1417                         FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1418                         FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1419 }
1420
1421 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1422 {
1423         if (stringset == ETH_SS_STATS)
1424                 memcpy(data, stats_strings, sizeof(stats_strings));
1425 }
1426
1427 /*
1428  * port stats maintained per queue of the port.  They should be in the same
1429  * order as in stats_strings above.
1430  */
1431 struct queue_port_stats {
1432         u64 tso;
1433         u64 tx_csum;
1434         u64 rx_csum;
1435         u64 vlan_ex;
1436         u64 vlan_ins;
1437         u64 gro_pkts;
1438         u64 gro_merged;
1439 };
1440
1441 static void collect_sge_port_stats(const struct adapter *adap,
1442                 const struct port_info *p, struct queue_port_stats *s)
1443 {
1444         int i;
1445         const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1446         const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1447
1448         memset(s, 0, sizeof(*s));
1449         for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1450                 s->tso += tx->tso;
1451                 s->tx_csum += tx->tx_cso;
1452                 s->rx_csum += rx->stats.rx_cso;
1453                 s->vlan_ex += rx->stats.vlan_ex;
1454                 s->vlan_ins += tx->vlan_ins;
1455                 s->gro_pkts += rx->stats.lro_pkts;
1456                 s->gro_merged += rx->stats.lro_merged;
1457         }
1458 }
1459
1460 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1461                       u64 *data)
1462 {
1463         struct port_info *pi = netdev_priv(dev);
1464         struct adapter *adapter = pi->adapter;
1465         u32 val1, val2;
1466
1467         t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1468
1469         data += sizeof(struct port_stats) / sizeof(u64);
1470         collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1471         data += sizeof(struct queue_port_stats) / sizeof(u64);
1472         if (!is_t4(adapter->chip)) {
1473                 t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1474                 val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1475                 val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1476                 *data = val1 - val2;
1477                 data++;
1478                 *data = val2;
1479                 data++;
1480         } else {
1481                 memset(data, 0, 2 * sizeof(u64));
1482                 *data += 2;
1483         }
1484 }
1485
1486 /*
1487  * Return a version number to identify the type of adapter.  The scheme is:
1488  * - bits 0..9: chip version
1489  * - bits 10..15: chip revision
1490  * - bits 16..23: register dump version
1491  */
1492 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1493 {
1494         return CHELSIO_CHIP_VERSION(ap->chip) |
1495                 (CHELSIO_CHIP_RELEASE(ap->chip) << 10) | (1 << 16);
1496 }
1497
1498 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1499                            unsigned int end)
1500 {
1501         u32 *p = buf + start;
1502
1503         for ( ; start <= end; start += sizeof(u32))
1504                 *p++ = t4_read_reg(ap, start);
1505 }
1506
1507 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1508                      void *buf)
1509 {
1510         static const unsigned int t4_reg_ranges[] = {
1511                 0x1008, 0x1108,
1512                 0x1180, 0x11b4,
1513                 0x11fc, 0x123c,
1514                 0x1300, 0x173c,
1515                 0x1800, 0x18fc,
1516                 0x3000, 0x30d8,
1517                 0x30e0, 0x5924,
1518                 0x5960, 0x59d4,
1519                 0x5a00, 0x5af8,
1520                 0x6000, 0x6098,
1521                 0x6100, 0x6150,
1522                 0x6200, 0x6208,
1523                 0x6240, 0x6248,
1524                 0x6280, 0x6338,
1525                 0x6370, 0x638c,
1526                 0x6400, 0x643c,
1527                 0x6500, 0x6524,
1528                 0x6a00, 0x6a38,
1529                 0x6a60, 0x6a78,
1530                 0x6b00, 0x6b84,
1531                 0x6bf0, 0x6c84,
1532                 0x6cf0, 0x6d84,
1533                 0x6df0, 0x6e84,
1534                 0x6ef0, 0x6f84,
1535                 0x6ff0, 0x7084,
1536                 0x70f0, 0x7184,
1537                 0x71f0, 0x7284,
1538                 0x72f0, 0x7384,
1539                 0x73f0, 0x7450,
1540                 0x7500, 0x7530,
1541                 0x7600, 0x761c,
1542                 0x7680, 0x76cc,
1543                 0x7700, 0x7798,
1544                 0x77c0, 0x77fc,
1545                 0x7900, 0x79fc,
1546                 0x7b00, 0x7c38,
1547                 0x7d00, 0x7efc,
1548                 0x8dc0, 0x8e1c,
1549                 0x8e30, 0x8e78,
1550                 0x8ea0, 0x8f6c,
1551                 0x8fc0, 0x9074,
1552                 0x90fc, 0x90fc,
1553                 0x9400, 0x9458,
1554                 0x9600, 0x96bc,
1555                 0x9800, 0x9808,
1556                 0x9820, 0x983c,
1557                 0x9850, 0x9864,
1558                 0x9c00, 0x9c6c,
1559                 0x9c80, 0x9cec,
1560                 0x9d00, 0x9d6c,
1561                 0x9d80, 0x9dec,
1562                 0x9e00, 0x9e6c,
1563                 0x9e80, 0x9eec,
1564                 0x9f00, 0x9f6c,
1565                 0x9f80, 0x9fec,
1566                 0xd004, 0xd03c,
1567                 0xdfc0, 0xdfe0,
1568                 0xe000, 0xea7c,
1569                 0xf000, 0x11190,
1570                 0x19040, 0x1906c,
1571                 0x19078, 0x19080,
1572                 0x1908c, 0x19124,
1573                 0x19150, 0x191b0,
1574                 0x191d0, 0x191e8,
1575                 0x19238, 0x1924c,
1576                 0x193f8, 0x19474,
1577                 0x19490, 0x194f8,
1578                 0x19800, 0x19f30,
1579                 0x1a000, 0x1a06c,
1580                 0x1a0b0, 0x1a120,
1581                 0x1a128, 0x1a138,
1582                 0x1a190, 0x1a1c4,
1583                 0x1a1fc, 0x1a1fc,
1584                 0x1e040, 0x1e04c,
1585                 0x1e284, 0x1e28c,
1586                 0x1e2c0, 0x1e2c0,
1587                 0x1e2e0, 0x1e2e0,
1588                 0x1e300, 0x1e384,
1589                 0x1e3c0, 0x1e3c8,
1590                 0x1e440, 0x1e44c,
1591                 0x1e684, 0x1e68c,
1592                 0x1e6c0, 0x1e6c0,
1593                 0x1e6e0, 0x1e6e0,
1594                 0x1e700, 0x1e784,
1595                 0x1e7c0, 0x1e7c8,
1596                 0x1e840, 0x1e84c,
1597                 0x1ea84, 0x1ea8c,
1598                 0x1eac0, 0x1eac0,
1599                 0x1eae0, 0x1eae0,
1600                 0x1eb00, 0x1eb84,
1601                 0x1ebc0, 0x1ebc8,
1602                 0x1ec40, 0x1ec4c,
1603                 0x1ee84, 0x1ee8c,
1604                 0x1eec0, 0x1eec0,
1605                 0x1eee0, 0x1eee0,
1606                 0x1ef00, 0x1ef84,
1607                 0x1efc0, 0x1efc8,
1608                 0x1f040, 0x1f04c,
1609                 0x1f284, 0x1f28c,
1610                 0x1f2c0, 0x1f2c0,
1611                 0x1f2e0, 0x1f2e0,
1612                 0x1f300, 0x1f384,
1613                 0x1f3c0, 0x1f3c8,
1614                 0x1f440, 0x1f44c,
1615                 0x1f684, 0x1f68c,
1616                 0x1f6c0, 0x1f6c0,
1617                 0x1f6e0, 0x1f6e0,
1618                 0x1f700, 0x1f784,
1619                 0x1f7c0, 0x1f7c8,
1620                 0x1f840, 0x1f84c,
1621                 0x1fa84, 0x1fa8c,
1622                 0x1fac0, 0x1fac0,
1623                 0x1fae0, 0x1fae0,
1624                 0x1fb00, 0x1fb84,
1625                 0x1fbc0, 0x1fbc8,
1626                 0x1fc40, 0x1fc4c,
1627                 0x1fe84, 0x1fe8c,
1628                 0x1fec0, 0x1fec0,
1629                 0x1fee0, 0x1fee0,
1630                 0x1ff00, 0x1ff84,
1631                 0x1ffc0, 0x1ffc8,
1632                 0x20000, 0x2002c,
1633                 0x20100, 0x2013c,
1634                 0x20190, 0x201c8,
1635                 0x20200, 0x20318,
1636                 0x20400, 0x20528,
1637                 0x20540, 0x20614,
1638                 0x21000, 0x21040,
1639                 0x2104c, 0x21060,
1640                 0x210c0, 0x210ec,
1641                 0x21200, 0x21268,
1642                 0x21270, 0x21284,
1643                 0x212fc, 0x21388,
1644                 0x21400, 0x21404,
1645                 0x21500, 0x21518,
1646                 0x2152c, 0x2153c,
1647                 0x21550, 0x21554,
1648                 0x21600, 0x21600,
1649                 0x21608, 0x21628,
1650                 0x21630, 0x2163c,
1651                 0x21700, 0x2171c,
1652                 0x21780, 0x2178c,
1653                 0x21800, 0x21c38,
1654                 0x21c80, 0x21d7c,
1655                 0x21e00, 0x21e04,
1656                 0x22000, 0x2202c,
1657                 0x22100, 0x2213c,
1658                 0x22190, 0x221c8,
1659                 0x22200, 0x22318,
1660                 0x22400, 0x22528,
1661                 0x22540, 0x22614,
1662                 0x23000, 0x23040,
1663                 0x2304c, 0x23060,
1664                 0x230c0, 0x230ec,
1665                 0x23200, 0x23268,
1666                 0x23270, 0x23284,
1667                 0x232fc, 0x23388,
1668                 0x23400, 0x23404,
1669                 0x23500, 0x23518,
1670                 0x2352c, 0x2353c,
1671                 0x23550, 0x23554,
1672                 0x23600, 0x23600,
1673                 0x23608, 0x23628,
1674                 0x23630, 0x2363c,
1675                 0x23700, 0x2371c,
1676                 0x23780, 0x2378c,
1677                 0x23800, 0x23c38,
1678                 0x23c80, 0x23d7c,
1679                 0x23e00, 0x23e04,
1680                 0x24000, 0x2402c,
1681                 0x24100, 0x2413c,
1682                 0x24190, 0x241c8,
1683                 0x24200, 0x24318,
1684                 0x24400, 0x24528,
1685                 0x24540, 0x24614,
1686                 0x25000, 0x25040,
1687                 0x2504c, 0x25060,
1688                 0x250c0, 0x250ec,
1689                 0x25200, 0x25268,
1690                 0x25270, 0x25284,
1691                 0x252fc, 0x25388,
1692                 0x25400, 0x25404,
1693                 0x25500, 0x25518,
1694                 0x2552c, 0x2553c,
1695                 0x25550, 0x25554,
1696                 0x25600, 0x25600,
1697                 0x25608, 0x25628,
1698                 0x25630, 0x2563c,
1699                 0x25700, 0x2571c,
1700                 0x25780, 0x2578c,
1701                 0x25800, 0x25c38,
1702                 0x25c80, 0x25d7c,
1703                 0x25e00, 0x25e04,
1704                 0x26000, 0x2602c,
1705                 0x26100, 0x2613c,
1706                 0x26190, 0x261c8,
1707                 0x26200, 0x26318,
1708                 0x26400, 0x26528,
1709                 0x26540, 0x26614,
1710                 0x27000, 0x27040,
1711                 0x2704c, 0x27060,
1712                 0x270c0, 0x270ec,
1713                 0x27200, 0x27268,
1714                 0x27270, 0x27284,
1715                 0x272fc, 0x27388,
1716                 0x27400, 0x27404,
1717                 0x27500, 0x27518,
1718                 0x2752c, 0x2753c,
1719                 0x27550, 0x27554,
1720                 0x27600, 0x27600,
1721                 0x27608, 0x27628,
1722                 0x27630, 0x2763c,
1723                 0x27700, 0x2771c,
1724                 0x27780, 0x2778c,
1725                 0x27800, 0x27c38,
1726                 0x27c80, 0x27d7c,
1727                 0x27e00, 0x27e04
1728         };
1729
1730         static const unsigned int t5_reg_ranges[] = {
1731                 0x1008, 0x1148,
1732                 0x1180, 0x11b4,
1733                 0x11fc, 0x123c,
1734                 0x1280, 0x173c,
1735                 0x1800, 0x18fc,
1736                 0x3000, 0x3028,
1737                 0x3060, 0x30d8,
1738                 0x30e0, 0x30fc,
1739                 0x3140, 0x357c,
1740                 0x35a8, 0x35cc,
1741                 0x35ec, 0x35ec,
1742                 0x3600, 0x5624,
1743                 0x56cc, 0x575c,
1744                 0x580c, 0x5814,
1745                 0x5890, 0x58bc,
1746                 0x5940, 0x59dc,
1747                 0x59fc, 0x5a18,
1748                 0x5a60, 0x5a9c,
1749                 0x5b9c, 0x5bfc,
1750                 0x6000, 0x6040,
1751                 0x6058, 0x614c,
1752                 0x7700, 0x7798,
1753                 0x77c0, 0x78fc,
1754                 0x7b00, 0x7c54,
1755                 0x7d00, 0x7efc,
1756                 0x8dc0, 0x8de0,
1757                 0x8df8, 0x8e84,
1758                 0x8ea0, 0x8f84,
1759                 0x8fc0, 0x90f8,
1760                 0x9400, 0x9470,
1761                 0x9600, 0x96f4,
1762                 0x9800, 0x9808,
1763                 0x9820, 0x983c,
1764                 0x9850, 0x9864,
1765                 0x9c00, 0x9c6c,
1766                 0x9c80, 0x9cec,
1767                 0x9d00, 0x9d6c,
1768                 0x9d80, 0x9dec,
1769                 0x9e00, 0x9e6c,
1770                 0x9e80, 0x9eec,
1771                 0x9f00, 0x9f6c,
1772                 0x9f80, 0xa020,
1773                 0xd004, 0xd03c,
1774                 0xdfc0, 0xdfe0,
1775                 0xe000, 0x11088,
1776                 0x1109c, 0x1117c,
1777                 0x11190, 0x11204,
1778                 0x19040, 0x1906c,
1779                 0x19078, 0x19080,
1780                 0x1908c, 0x19124,
1781                 0x19150, 0x191b0,
1782                 0x191d0, 0x191e8,
1783                 0x19238, 0x19290,
1784                 0x193f8, 0x19474,
1785                 0x19490, 0x194cc,
1786                 0x194f0, 0x194f8,
1787                 0x19c00, 0x19c60,
1788                 0x19c94, 0x19e10,
1789                 0x19e50, 0x19f34,
1790                 0x19f40, 0x19f50,
1791                 0x19f90, 0x19fe4,
1792                 0x1a000, 0x1a06c,
1793                 0x1a0b0, 0x1a120,
1794                 0x1a128, 0x1a138,
1795                 0x1a190, 0x1a1c4,
1796                 0x1a1fc, 0x1a1fc,
1797                 0x1e008, 0x1e00c,
1798                 0x1e040, 0x1e04c,
1799                 0x1e284, 0x1e290,
1800                 0x1e2c0, 0x1e2c0,
1801                 0x1e2e0, 0x1e2e0,
1802                 0x1e300, 0x1e384,
1803                 0x1e3c0, 0x1e3c8,
1804                 0x1e408, 0x1e40c,
1805                 0x1e440, 0x1e44c,
1806                 0x1e684, 0x1e690,
1807                 0x1e6c0, 0x1e6c0,
1808                 0x1e6e0, 0x1e6e0,
1809                 0x1e700, 0x1e784,
1810                 0x1e7c0, 0x1e7c8,
1811                 0x1e808, 0x1e80c,
1812                 0x1e840, 0x1e84c,
1813                 0x1ea84, 0x1ea90,
1814                 0x1eac0, 0x1eac0,
1815                 0x1eae0, 0x1eae0,
1816                 0x1eb00, 0x1eb84,
1817                 0x1ebc0, 0x1ebc8,
1818                 0x1ec08, 0x1ec0c,
1819                 0x1ec40, 0x1ec4c,
1820                 0x1ee84, 0x1ee90,
1821                 0x1eec0, 0x1eec0,
1822                 0x1eee0, 0x1eee0,
1823                 0x1ef00, 0x1ef84,
1824                 0x1efc0, 0x1efc8,
1825                 0x1f008, 0x1f00c,
1826                 0x1f040, 0x1f04c,
1827                 0x1f284, 0x1f290,
1828                 0x1f2c0, 0x1f2c0,
1829                 0x1f2e0, 0x1f2e0,
1830                 0x1f300, 0x1f384,
1831                 0x1f3c0, 0x1f3c8,
1832                 0x1f408, 0x1f40c,
1833                 0x1f440, 0x1f44c,
1834                 0x1f684, 0x1f690,
1835                 0x1f6c0, 0x1f6c0,
1836                 0x1f6e0, 0x1f6e0,
1837                 0x1f700, 0x1f784,
1838                 0x1f7c0, 0x1f7c8,
1839                 0x1f808, 0x1f80c,
1840                 0x1f840, 0x1f84c,
1841                 0x1fa84, 0x1fa90,
1842                 0x1fac0, 0x1fac0,
1843                 0x1fae0, 0x1fae0,
1844                 0x1fb00, 0x1fb84,
1845                 0x1fbc0, 0x1fbc8,
1846                 0x1fc08, 0x1fc0c,
1847                 0x1fc40, 0x1fc4c,
1848                 0x1fe84, 0x1fe90,
1849                 0x1fec0, 0x1fec0,
1850                 0x1fee0, 0x1fee0,
1851                 0x1ff00, 0x1ff84,
1852                 0x1ffc0, 0x1ffc8,
1853                 0x30000, 0x30030,
1854                 0x30100, 0x30144,
1855                 0x30190, 0x301d0,
1856                 0x30200, 0x30318,
1857                 0x30400, 0x3052c,
1858                 0x30540, 0x3061c,
1859                 0x30800, 0x30834,
1860                 0x308c0, 0x30908,
1861                 0x30910, 0x309ac,
1862                 0x30a00, 0x30a04,
1863                 0x30a0c, 0x30a2c,
1864                 0x30a44, 0x30a50,
1865                 0x30a74, 0x30c24,
1866                 0x30d08, 0x30d14,
1867                 0x30d1c, 0x30d20,
1868                 0x30d3c, 0x30d50,
1869                 0x31200, 0x3120c,
1870                 0x31220, 0x31220,
1871                 0x31240, 0x31240,
1872                 0x31600, 0x31600,
1873                 0x31608, 0x3160c,
1874                 0x31a00, 0x31a1c,
1875                 0x31e04, 0x31e20,
1876                 0x31e38, 0x31e3c,
1877                 0x31e80, 0x31e80,
1878                 0x31e88, 0x31ea8,
1879                 0x31eb0, 0x31eb4,
1880                 0x31ec8, 0x31ed4,
1881                 0x31fb8, 0x32004,
1882                 0x32208, 0x3223c,
1883                 0x32600, 0x32630,
1884                 0x32a00, 0x32abc,
1885                 0x32b00, 0x32b70,
1886                 0x33000, 0x33048,
1887                 0x33060, 0x3309c,
1888                 0x330f0, 0x33148,
1889                 0x33160, 0x3319c,
1890                 0x331f0, 0x332e4,
1891                 0x332f8, 0x333e4,
1892                 0x333f8, 0x33448,
1893                 0x33460, 0x3349c,
1894                 0x334f0, 0x33548,
1895                 0x33560, 0x3359c,
1896                 0x335f0, 0x336e4,
1897                 0x336f8, 0x337e4,
1898                 0x337f8, 0x337fc,
1899                 0x33814, 0x33814,
1900                 0x3382c, 0x3382c,
1901                 0x33880, 0x3388c,
1902                 0x338e8, 0x338ec,
1903                 0x33900, 0x33948,
1904                 0x33960, 0x3399c,
1905                 0x339f0, 0x33ae4,
1906                 0x33af8, 0x33b10,
1907                 0x33b28, 0x33b28,
1908                 0x33b3c, 0x33b50,
1909                 0x33bf0, 0x33c10,
1910                 0x33c28, 0x33c28,
1911                 0x33c3c, 0x33c50,
1912                 0x33cf0, 0x33cfc,
1913                 0x34000, 0x34030,
1914                 0x34100, 0x34144,
1915                 0x34190, 0x341d0,
1916                 0x34200, 0x34318,
1917                 0x34400, 0x3452c,
1918                 0x34540, 0x3461c,
1919                 0x34800, 0x34834,
1920                 0x348c0, 0x34908,
1921                 0x34910, 0x349ac,
1922                 0x34a00, 0x34a04,
1923                 0x34a0c, 0x34a2c,
1924                 0x34a44, 0x34a50,
1925                 0x34a74, 0x34c24,
1926                 0x34d08, 0x34d14,
1927                 0x34d1c, 0x34d20,
1928                 0x34d3c, 0x34d50,
1929                 0x35200, 0x3520c,
1930                 0x35220, 0x35220,
1931                 0x35240, 0x35240,
1932                 0x35600, 0x35600,
1933                 0x35608, 0x3560c,
1934                 0x35a00, 0x35a1c,
1935                 0x35e04, 0x35e20,
1936                 0x35e38, 0x35e3c,
1937                 0x35e80, 0x35e80,
1938                 0x35e88, 0x35ea8,
1939                 0x35eb0, 0x35eb4,
1940                 0x35ec8, 0x35ed4,
1941                 0x35fb8, 0x36004,
1942                 0x36208, 0x3623c,
1943                 0x36600, 0x36630,
1944                 0x36a00, 0x36abc,
1945                 0x36b00, 0x36b70,
1946                 0x37000, 0x37048,
1947                 0x37060, 0x3709c,
1948                 0x370f0, 0x37148,
1949                 0x37160, 0x3719c,
1950                 0x371f0, 0x372e4,
1951                 0x372f8, 0x373e4,
1952                 0x373f8, 0x37448,
1953                 0x37460, 0x3749c,
1954                 0x374f0, 0x37548,
1955                 0x37560, 0x3759c,
1956                 0x375f0, 0x376e4,
1957                 0x376f8, 0x377e4,
1958                 0x377f8, 0x377fc,
1959                 0x37814, 0x37814,
1960                 0x3782c, 0x3782c,
1961                 0x37880, 0x3788c,
1962                 0x378e8, 0x378ec,
1963                 0x37900, 0x37948,
1964                 0x37960, 0x3799c,
1965                 0x379f0, 0x37ae4,
1966                 0x37af8, 0x37b10,
1967                 0x37b28, 0x37b28,
1968                 0x37b3c, 0x37b50,
1969                 0x37bf0, 0x37c10,
1970                 0x37c28, 0x37c28,
1971                 0x37c3c, 0x37c50,
1972                 0x37cf0, 0x37cfc,
1973                 0x38000, 0x38030,
1974                 0x38100, 0x38144,
1975                 0x38190, 0x381d0,
1976                 0x38200, 0x38318,
1977                 0x38400, 0x3852c,
1978                 0x38540, 0x3861c,
1979                 0x38800, 0x38834,
1980                 0x388c0, 0x38908,
1981                 0x38910, 0x389ac,
1982                 0x38a00, 0x38a04,
1983                 0x38a0c, 0x38a2c,
1984                 0x38a44, 0x38a50,
1985                 0x38a74, 0x38c24,
1986                 0x38d08, 0x38d14,
1987                 0x38d1c, 0x38d20,
1988                 0x38d3c, 0x38d50,
1989                 0x39200, 0x3920c,
1990                 0x39220, 0x39220,
1991                 0x39240, 0x39240,
1992                 0x39600, 0x39600,
1993                 0x39608, 0x3960c,
1994                 0x39a00, 0x39a1c,
1995                 0x39e04, 0x39e20,
1996                 0x39e38, 0x39e3c,
1997                 0x39e80, 0x39e80,
1998                 0x39e88, 0x39ea8,
1999                 0x39eb0, 0x39eb4,
2000                 0x39ec8, 0x39ed4,
2001                 0x39fb8, 0x3a004,
2002                 0x3a208, 0x3a23c,
2003                 0x3a600, 0x3a630,
2004                 0x3aa00, 0x3aabc,
2005                 0x3ab00, 0x3ab70,
2006                 0x3b000, 0x3b048,
2007                 0x3b060, 0x3b09c,
2008                 0x3b0f0, 0x3b148,
2009                 0x3b160, 0x3b19c,
2010                 0x3b1f0, 0x3b2e4,
2011                 0x3b2f8, 0x3b3e4,
2012                 0x3b3f8, 0x3b448,
2013                 0x3b460, 0x3b49c,
2014                 0x3b4f0, 0x3b548,
2015                 0x3b560, 0x3b59c,
2016                 0x3b5f0, 0x3b6e4,
2017                 0x3b6f8, 0x3b7e4,
2018                 0x3b7f8, 0x3b7fc,
2019                 0x3b814, 0x3b814,
2020                 0x3b82c, 0x3b82c,
2021                 0x3b880, 0x3b88c,
2022                 0x3b8e8, 0x3b8ec,
2023                 0x3b900, 0x3b948,
2024                 0x3b960, 0x3b99c,
2025                 0x3b9f0, 0x3bae4,
2026                 0x3baf8, 0x3bb10,
2027                 0x3bb28, 0x3bb28,
2028                 0x3bb3c, 0x3bb50,
2029                 0x3bbf0, 0x3bc10,
2030                 0x3bc28, 0x3bc28,
2031                 0x3bc3c, 0x3bc50,
2032                 0x3bcf0, 0x3bcfc,
2033                 0x3c000, 0x3c030,
2034                 0x3c100, 0x3c144,
2035                 0x3c190, 0x3c1d0,
2036                 0x3c200, 0x3c318,
2037                 0x3c400, 0x3c52c,
2038                 0x3c540, 0x3c61c,
2039                 0x3c800, 0x3c834,
2040                 0x3c8c0, 0x3c908,
2041                 0x3c910, 0x3c9ac,
2042                 0x3ca00, 0x3ca04,
2043                 0x3ca0c, 0x3ca2c,
2044                 0x3ca44, 0x3ca50,
2045                 0x3ca74, 0x3cc24,
2046                 0x3cd08, 0x3cd14,
2047                 0x3cd1c, 0x3cd20,
2048                 0x3cd3c, 0x3cd50,
2049                 0x3d200, 0x3d20c,
2050                 0x3d220, 0x3d220,
2051                 0x3d240, 0x3d240,
2052                 0x3d600, 0x3d600,
2053                 0x3d608, 0x3d60c,
2054                 0x3da00, 0x3da1c,
2055                 0x3de04, 0x3de20,
2056                 0x3de38, 0x3de3c,
2057                 0x3de80, 0x3de80,
2058                 0x3de88, 0x3dea8,
2059                 0x3deb0, 0x3deb4,
2060                 0x3dec8, 0x3ded4,
2061                 0x3dfb8, 0x3e004,
2062                 0x3e208, 0x3e23c,
2063                 0x3e600, 0x3e630,
2064                 0x3ea00, 0x3eabc,
2065                 0x3eb00, 0x3eb70,
2066                 0x3f000, 0x3f048,
2067                 0x3f060, 0x3f09c,
2068                 0x3f0f0, 0x3f148,
2069                 0x3f160, 0x3f19c,
2070                 0x3f1f0, 0x3f2e4,
2071                 0x3f2f8, 0x3f3e4,
2072                 0x3f3f8, 0x3f448,
2073                 0x3f460, 0x3f49c,
2074                 0x3f4f0, 0x3f548,
2075                 0x3f560, 0x3f59c,
2076                 0x3f5f0, 0x3f6e4,
2077                 0x3f6f8, 0x3f7e4,
2078                 0x3f7f8, 0x3f7fc,
2079                 0x3f814, 0x3f814,
2080                 0x3f82c, 0x3f82c,
2081                 0x3f880, 0x3f88c,
2082                 0x3f8e8, 0x3f8ec,
2083                 0x3f900, 0x3f948,
2084                 0x3f960, 0x3f99c,
2085                 0x3f9f0, 0x3fae4,
2086                 0x3faf8, 0x3fb10,
2087                 0x3fb28, 0x3fb28,
2088                 0x3fb3c, 0x3fb50,
2089                 0x3fbf0, 0x3fc10,
2090                 0x3fc28, 0x3fc28,
2091                 0x3fc3c, 0x3fc50,
2092                 0x3fcf0, 0x3fcfc,
2093                 0x40000, 0x4000c,
2094                 0x40040, 0x40068,
2095                 0x40080, 0x40144,
2096                 0x40180, 0x4018c,
2097                 0x40200, 0x40298,
2098                 0x402ac, 0x4033c,
2099                 0x403f8, 0x403fc,
2100                 0x41300, 0x413c4,
2101                 0x41400, 0x4141c,
2102                 0x41480, 0x414d0,
2103                 0x44000, 0x44078,
2104                 0x440c0, 0x44278,
2105                 0x442c0, 0x44478,
2106                 0x444c0, 0x44678,
2107                 0x446c0, 0x44878,
2108                 0x448c0, 0x449fc,
2109                 0x45000, 0x45068,
2110                 0x45080, 0x45084,
2111                 0x450a0, 0x450b0,
2112                 0x45200, 0x45268,
2113                 0x45280, 0x45284,
2114                 0x452a0, 0x452b0,
2115                 0x460c0, 0x460e4,
2116                 0x47000, 0x4708c,
2117                 0x47200, 0x47250,
2118                 0x47400, 0x47420,
2119                 0x47600, 0x47618,
2120                 0x47800, 0x47814,
2121                 0x48000, 0x4800c,
2122                 0x48040, 0x48068,
2123                 0x48080, 0x48144,
2124                 0x48180, 0x4818c,
2125                 0x48200, 0x48298,
2126                 0x482ac, 0x4833c,
2127                 0x483f8, 0x483fc,
2128                 0x49300, 0x493c4,
2129                 0x49400, 0x4941c,
2130                 0x49480, 0x494d0,
2131                 0x4c000, 0x4c078,
2132                 0x4c0c0, 0x4c278,
2133                 0x4c2c0, 0x4c478,
2134                 0x4c4c0, 0x4c678,
2135                 0x4c6c0, 0x4c878,
2136                 0x4c8c0, 0x4c9fc,
2137                 0x4d000, 0x4d068,
2138                 0x4d080, 0x4d084,
2139                 0x4d0a0, 0x4d0b0,
2140                 0x4d200, 0x4d268,
2141                 0x4d280, 0x4d284,
2142                 0x4d2a0, 0x4d2b0,
2143                 0x4e0c0, 0x4e0e4,
2144                 0x4f000, 0x4f08c,
2145                 0x4f200, 0x4f250,
2146                 0x4f400, 0x4f420,
2147                 0x4f600, 0x4f618,
2148                 0x4f800, 0x4f814,
2149                 0x50000, 0x500cc,
2150                 0x50400, 0x50400,
2151                 0x50800, 0x508cc,
2152                 0x50c00, 0x50c00,
2153                 0x51000, 0x5101c,
2154                 0x51300, 0x51308,
2155         };
2156
2157         int i;
2158         struct adapter *ap = netdev2adap(dev);
2159         static const unsigned int *reg_ranges;
2160         int arr_size = 0, buf_size = 0;
2161
2162         if (is_t4(ap->chip)) {
2163                 reg_ranges = &t4_reg_ranges[0];
2164                 arr_size = ARRAY_SIZE(t4_reg_ranges);
2165                 buf_size = T4_REGMAP_SIZE;
2166         } else {
2167                 reg_ranges = &t5_reg_ranges[0];
2168                 arr_size = ARRAY_SIZE(t5_reg_ranges);
2169                 buf_size = T5_REGMAP_SIZE;
2170         }
2171
2172         regs->version = mk_adap_vers(ap);
2173
2174         memset(buf, 0, buf_size);
2175         for (i = 0; i < arr_size; i += 2)
2176                 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2177 }
2178
2179 static int restart_autoneg(struct net_device *dev)
2180 {
2181         struct port_info *p = netdev_priv(dev);
2182
2183         if (!netif_running(dev))
2184                 return -EAGAIN;
2185         if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2186                 return -EINVAL;
2187         t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2188         return 0;
2189 }
2190
2191 static int identify_port(struct net_device *dev,
2192                          enum ethtool_phys_id_state state)
2193 {
2194         unsigned int val;
2195         struct adapter *adap = netdev2adap(dev);
2196
2197         if (state == ETHTOOL_ID_ACTIVE)
2198                 val = 0xffff;
2199         else if (state == ETHTOOL_ID_INACTIVE)
2200                 val = 0;
2201         else
2202                 return -EINVAL;
2203
2204         return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2205 }
2206
2207 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2208 {
2209         unsigned int v = 0;
2210
2211         if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2212             type == FW_PORT_TYPE_BT_XAUI) {
2213                 v |= SUPPORTED_TP;
2214                 if (caps & FW_PORT_CAP_SPEED_100M)
2215                         v |= SUPPORTED_100baseT_Full;
2216                 if (caps & FW_PORT_CAP_SPEED_1G)
2217                         v |= SUPPORTED_1000baseT_Full;
2218                 if (caps & FW_PORT_CAP_SPEED_10G)
2219                         v |= SUPPORTED_10000baseT_Full;
2220         } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2221                 v |= SUPPORTED_Backplane;
2222                 if (caps & FW_PORT_CAP_SPEED_1G)
2223                         v |= SUPPORTED_1000baseKX_Full;
2224                 if (caps & FW_PORT_CAP_SPEED_10G)
2225                         v |= SUPPORTED_10000baseKX4_Full;
2226         } else if (type == FW_PORT_TYPE_KR)
2227                 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2228         else if (type == FW_PORT_TYPE_BP_AP)
2229                 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2230                      SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2231         else if (type == FW_PORT_TYPE_BP4_AP)
2232                 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2233                      SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2234                      SUPPORTED_10000baseKX4_Full;
2235         else if (type == FW_PORT_TYPE_FIBER_XFI ||
2236                  type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2237                 v |= SUPPORTED_FIBRE;
2238
2239         if (caps & FW_PORT_CAP_ANEG)
2240                 v |= SUPPORTED_Autoneg;
2241         return v;
2242 }
2243
2244 static unsigned int to_fw_linkcaps(unsigned int caps)
2245 {
2246         unsigned int v = 0;
2247
2248         if (caps & ADVERTISED_100baseT_Full)
2249                 v |= FW_PORT_CAP_SPEED_100M;
2250         if (caps & ADVERTISED_1000baseT_Full)
2251                 v |= FW_PORT_CAP_SPEED_1G;
2252         if (caps & ADVERTISED_10000baseT_Full)
2253                 v |= FW_PORT_CAP_SPEED_10G;
2254         return v;
2255 }
2256
2257 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2258 {
2259         const struct port_info *p = netdev_priv(dev);
2260
2261         if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2262             p->port_type == FW_PORT_TYPE_BT_XFI ||
2263             p->port_type == FW_PORT_TYPE_BT_XAUI)
2264                 cmd->port = PORT_TP;
2265         else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2266                  p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2267                 cmd->port = PORT_FIBRE;
2268         else if (p->port_type == FW_PORT_TYPE_SFP) {
2269                 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2270                     p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2271                         cmd->port = PORT_DA;
2272                 else
2273                         cmd->port = PORT_FIBRE;
2274         } else
2275                 cmd->port = PORT_OTHER;
2276
2277         if (p->mdio_addr >= 0) {
2278                 cmd->phy_address = p->mdio_addr;
2279                 cmd->transceiver = XCVR_EXTERNAL;
2280                 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2281                         MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2282         } else {
2283                 cmd->phy_address = 0;  /* not really, but no better option */
2284                 cmd->transceiver = XCVR_INTERNAL;
2285                 cmd->mdio_support = 0;
2286         }
2287
2288         cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2289         cmd->advertising = from_fw_linkcaps(p->port_type,
2290                                             p->link_cfg.advertising);
2291         ethtool_cmd_speed_set(cmd,
2292                               netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2293         cmd->duplex = DUPLEX_FULL;
2294         cmd->autoneg = p->link_cfg.autoneg;
2295         cmd->maxtxpkt = 0;
2296         cmd->maxrxpkt = 0;
2297         return 0;
2298 }
2299
2300 static unsigned int speed_to_caps(int speed)
2301 {
2302         if (speed == SPEED_100)
2303                 return FW_PORT_CAP_SPEED_100M;
2304         if (speed == SPEED_1000)
2305                 return FW_PORT_CAP_SPEED_1G;
2306         if (speed == SPEED_10000)
2307                 return FW_PORT_CAP_SPEED_10G;
2308         return 0;
2309 }
2310
2311 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2312 {
2313         unsigned int cap;
2314         struct port_info *p = netdev_priv(dev);
2315         struct link_config *lc = &p->link_cfg;
2316         u32 speed = ethtool_cmd_speed(cmd);
2317
2318         if (cmd->duplex != DUPLEX_FULL)     /* only full-duplex supported */
2319                 return -EINVAL;
2320
2321         if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2322                 /*
2323                  * PHY offers a single speed.  See if that's what's
2324                  * being requested.
2325                  */
2326                 if (cmd->autoneg == AUTONEG_DISABLE &&
2327                     (lc->supported & speed_to_caps(speed)))
2328                         return 0;
2329                 return -EINVAL;
2330         }
2331
2332         if (cmd->autoneg == AUTONEG_DISABLE) {
2333                 cap = speed_to_caps(speed);
2334
2335                 if (!(lc->supported & cap) || (speed == SPEED_1000) ||
2336                     (speed == SPEED_10000))
2337                         return -EINVAL;
2338                 lc->requested_speed = cap;
2339                 lc->advertising = 0;
2340         } else {
2341                 cap = to_fw_linkcaps(cmd->advertising);
2342                 if (!(lc->supported & cap))
2343                         return -EINVAL;
2344                 lc->requested_speed = 0;
2345                 lc->advertising = cap | FW_PORT_CAP_ANEG;
2346         }
2347         lc->autoneg = cmd->autoneg;
2348
2349         if (netif_running(dev))
2350                 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2351                                      lc);
2352         return 0;
2353 }
2354
2355 static void get_pauseparam(struct net_device *dev,
2356                            struct ethtool_pauseparam *epause)
2357 {
2358         struct port_info *p = netdev_priv(dev);
2359
2360         epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2361         epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2362         epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2363 }
2364
2365 static int set_pauseparam(struct net_device *dev,
2366                           struct ethtool_pauseparam *epause)
2367 {
2368         struct port_info *p = netdev_priv(dev);
2369         struct link_config *lc = &p->link_cfg;
2370
2371         if (epause->autoneg == AUTONEG_DISABLE)
2372                 lc->requested_fc = 0;
2373         else if (lc->supported & FW_PORT_CAP_ANEG)
2374                 lc->requested_fc = PAUSE_AUTONEG;
2375         else
2376                 return -EINVAL;
2377
2378         if (epause->rx_pause)
2379                 lc->requested_fc |= PAUSE_RX;
2380         if (epause->tx_pause)
2381                 lc->requested_fc |= PAUSE_TX;
2382         if (netif_running(dev))
2383                 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2384                                      lc);
2385         return 0;
2386 }
2387
2388 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2389 {
2390         const struct port_info *pi = netdev_priv(dev);
2391         const struct sge *s = &pi->adapter->sge;
2392
2393         e->rx_max_pending = MAX_RX_BUFFERS;
2394         e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2395         e->rx_jumbo_max_pending = 0;
2396         e->tx_max_pending = MAX_TXQ_ENTRIES;
2397
2398         e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2399         e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2400         e->rx_jumbo_pending = 0;
2401         e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2402 }
2403
2404 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2405 {
2406         int i;
2407         const struct port_info *pi = netdev_priv(dev);
2408         struct adapter *adapter = pi->adapter;
2409         struct sge *s = &adapter->sge;
2410
2411         if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2412             e->tx_pending > MAX_TXQ_ENTRIES ||
2413             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2414             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2415             e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2416                 return -EINVAL;
2417
2418         if (adapter->flags & FULL_INIT_DONE)
2419                 return -EBUSY;
2420
2421         for (i = 0; i < pi->nqsets; ++i) {
2422                 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2423                 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2424                 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2425         }
2426         return 0;
2427 }
2428
2429 static int closest_timer(const struct sge *s, int time)
2430 {
2431         int i, delta, match = 0, min_delta = INT_MAX;
2432
2433         for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2434                 delta = time - s->timer_val[i];
2435                 if (delta < 0)
2436                         delta = -delta;
2437                 if (delta < min_delta) {
2438                         min_delta = delta;
2439                         match = i;
2440                 }
2441         }
2442         return match;
2443 }
2444
2445 static int closest_thres(const struct sge *s, int thres)
2446 {
2447         int i, delta, match = 0, min_delta = INT_MAX;
2448
2449         for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2450                 delta = thres - s->counter_val[i];
2451                 if (delta < 0)
2452                         delta = -delta;
2453                 if (delta < min_delta) {
2454                         min_delta = delta;
2455                         match = i;
2456                 }
2457         }
2458         return match;
2459 }
2460
2461 /*
2462  * Return a queue's interrupt hold-off time in us.  0 means no timer.
2463  */
2464 static unsigned int qtimer_val(const struct adapter *adap,
2465                                const struct sge_rspq *q)
2466 {
2467         unsigned int idx = q->intr_params >> 1;
2468
2469         return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2470 }
2471
2472 /**
2473  *      set_rxq_intr_params - set a queue's interrupt holdoff parameters
2474  *      @adap: the adapter
2475  *      @q: the Rx queue
2476  *      @us: the hold-off time in us, or 0 to disable timer
2477  *      @cnt: the hold-off packet count, or 0 to disable counter
2478  *
2479  *      Sets an Rx queue's interrupt hold-off time and packet count.  At least
2480  *      one of the two needs to be enabled for the queue to generate interrupts.
2481  */
2482 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
2483                                unsigned int us, unsigned int cnt)
2484 {
2485         if ((us | cnt) == 0)
2486                 cnt = 1;
2487
2488         if (cnt) {
2489                 int err;
2490                 u32 v, new_idx;
2491
2492                 new_idx = closest_thres(&adap->sge, cnt);
2493                 if (q->desc && q->pktcnt_idx != new_idx) {
2494                         /* the queue has already been created, update it */
2495                         v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2496                             FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2497                             FW_PARAMS_PARAM_YZ(q->cntxt_id);
2498                         err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2499                                             &new_idx);
2500                         if (err)
2501                                 return err;
2502                 }
2503                 q->pktcnt_idx = new_idx;
2504         }
2505
2506         us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2507         q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2508         return 0;
2509 }
2510
2511 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2512 {
2513         const struct port_info *pi = netdev_priv(dev);
2514         struct adapter *adap = pi->adapter;
2515         struct sge_rspq *q;
2516         int i;
2517         int r = 0;
2518
2519         for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2520                 q = &adap->sge.ethrxq[i].rspq;
2521                 r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2522                         c->rx_max_coalesced_frames);
2523                 if (r) {
2524                         dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2525                         break;
2526                 }
2527         }
2528         return r;
2529 }
2530
2531 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2532 {
2533         const struct port_info *pi = netdev_priv(dev);
2534         const struct adapter *adap = pi->adapter;
2535         const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2536
2537         c->rx_coalesce_usecs = qtimer_val(adap, rq);
2538         c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2539                 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2540         return 0;
2541 }
2542
2543 /**
2544  *      eeprom_ptov - translate a physical EEPROM address to virtual
2545  *      @phys_addr: the physical EEPROM address
2546  *      @fn: the PCI function number
2547  *      @sz: size of function-specific area
2548  *
2549  *      Translate a physical EEPROM address to virtual.  The first 1K is
2550  *      accessed through virtual addresses starting at 31K, the rest is
2551  *      accessed through virtual addresses starting at 0.
2552  *
2553  *      The mapping is as follows:
2554  *      [0..1K) -> [31K..32K)
2555  *      [1K..1K+A) -> [31K-A..31K)
2556  *      [1K+A..ES) -> [0..ES-A-1K)
2557  *
2558  *      where A = @fn * @sz, and ES = EEPROM size.
2559  */
2560 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2561 {
2562         fn *= sz;
2563         if (phys_addr < 1024)
2564                 return phys_addr + (31 << 10);
2565         if (phys_addr < 1024 + fn)
2566                 return 31744 - fn + phys_addr - 1024;
2567         if (phys_addr < EEPROMSIZE)
2568                 return phys_addr - 1024 - fn;
2569         return -EINVAL;
2570 }
2571
2572 /*
2573  * The next two routines implement eeprom read/write from physical addresses.
2574  */
2575 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2576 {
2577         int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2578
2579         if (vaddr >= 0)
2580                 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2581         return vaddr < 0 ? vaddr : 0;
2582 }
2583
2584 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2585 {
2586         int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2587
2588         if (vaddr >= 0)
2589                 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2590         return vaddr < 0 ? vaddr : 0;
2591 }
2592
2593 #define EEPROM_MAGIC 0x38E2F10C
2594
2595 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2596                       u8 *data)
2597 {
2598         int i, err = 0;
2599         struct adapter *adapter = netdev2adap(dev);
2600
2601         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2602         if (!buf)
2603                 return -ENOMEM;
2604
2605         e->magic = EEPROM_MAGIC;
2606         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2607                 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2608
2609         if (!err)
2610                 memcpy(data, buf + e->offset, e->len);
2611         kfree(buf);
2612         return err;
2613 }
2614
2615 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2616                       u8 *data)
2617 {
2618         u8 *buf;
2619         int err = 0;
2620         u32 aligned_offset, aligned_len, *p;
2621         struct adapter *adapter = netdev2adap(dev);
2622
2623         if (eeprom->magic != EEPROM_MAGIC)
2624                 return -EINVAL;
2625
2626         aligned_offset = eeprom->offset & ~3;
2627         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2628
2629         if (adapter->fn > 0) {
2630                 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2631
2632                 if (aligned_offset < start ||
2633                     aligned_offset + aligned_len > start + EEPROMPFSIZE)
2634                         return -EPERM;
2635         }
2636
2637         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2638                 /*
2639                  * RMW possibly needed for first or last words.
2640                  */
2641                 buf = kmalloc(aligned_len, GFP_KERNEL);
2642                 if (!buf)
2643                         return -ENOMEM;
2644                 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2645                 if (!err && aligned_len > 4)
2646                         err = eeprom_rd_phys(adapter,
2647                                              aligned_offset + aligned_len - 4,
2648                                              (u32 *)&buf[aligned_len - 4]);
2649                 if (err)
2650                         goto out;
2651                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2652         } else
2653                 buf = data;
2654
2655         err = t4_seeprom_wp(adapter, false);
2656         if (err)
2657                 goto out;
2658
2659         for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2660                 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2661                 aligned_offset += 4;
2662         }
2663
2664         if (!err)
2665                 err = t4_seeprom_wp(adapter, true);
2666 out:
2667         if (buf != data)
2668                 kfree(buf);
2669         return err;
2670 }
2671
2672 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2673 {
2674         int ret;
2675         const struct firmware *fw;
2676         struct adapter *adap = netdev2adap(netdev);
2677
2678         ef->data[sizeof(ef->data) - 1] = '\0';
2679         ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2680         if (ret < 0)
2681                 return ret;
2682
2683         ret = t4_load_fw(adap, fw->data, fw->size);
2684         release_firmware(fw);
2685         if (!ret)
2686                 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2687         return ret;
2688 }
2689
2690 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2691 #define BCAST_CRC 0xa0ccc1a6
2692
2693 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2694 {
2695         wol->supported = WAKE_BCAST | WAKE_MAGIC;
2696         wol->wolopts = netdev2adap(dev)->wol;
2697         memset(&wol->sopass, 0, sizeof(wol->sopass));
2698 }
2699
2700 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2701 {
2702         int err = 0;
2703         struct port_info *pi = netdev_priv(dev);
2704
2705         if (wol->wolopts & ~WOL_SUPPORTED)
2706                 return -EINVAL;
2707         t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2708                             (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2709         if (wol->wolopts & WAKE_BCAST) {
2710                 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2711                                         ~0ULL, 0, false);
2712                 if (!err)
2713                         err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2714                                                 ~6ULL, ~0ULL, BCAST_CRC, true);
2715         } else
2716                 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2717         return err;
2718 }
2719
2720 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2721 {
2722         const struct port_info *pi = netdev_priv(dev);
2723         netdev_features_t changed = dev->features ^ features;
2724         int err;
2725
2726         if (!(changed & NETIF_F_HW_VLAN_RX))
2727                 return 0;
2728
2729         err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2730                             -1, -1, -1,
2731                             !!(features & NETIF_F_HW_VLAN_RX), true);
2732         if (unlikely(err))
2733                 dev->features = features ^ NETIF_F_HW_VLAN_RX;
2734         return err;
2735 }
2736
2737 static u32 get_rss_table_size(struct net_device *dev)
2738 {
2739         const struct port_info *pi = netdev_priv(dev);
2740
2741         return pi->rss_size;
2742 }
2743
2744 static int get_rss_table(struct net_device *dev, u32 *p)
2745 {
2746         const struct port_info *pi = netdev_priv(dev);
2747         unsigned int n = pi->rss_size;
2748
2749         while (n--)
2750                 p[n] = pi->rss[n];
2751         return 0;
2752 }
2753
2754 static int set_rss_table(struct net_device *dev, const u32 *p)
2755 {
2756         unsigned int i;
2757         struct port_info *pi = netdev_priv(dev);
2758
2759         for (i = 0; i < pi->rss_size; i++)
2760                 pi->rss[i] = p[i];
2761         if (pi->adapter->flags & FULL_INIT_DONE)
2762                 return write_rss(pi, pi->rss);
2763         return 0;
2764 }
2765
2766 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2767                      u32 *rules)
2768 {
2769         const struct port_info *pi = netdev_priv(dev);
2770
2771         switch (info->cmd) {
2772         case ETHTOOL_GRXFH: {
2773                 unsigned int v = pi->rss_mode;
2774
2775                 info->data = 0;
2776                 switch (info->flow_type) {
2777                 case TCP_V4_FLOW:
2778                         if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2779                                 info->data = RXH_IP_SRC | RXH_IP_DST |
2780                                              RXH_L4_B_0_1 | RXH_L4_B_2_3;
2781                         else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2782                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2783                         break;
2784                 case UDP_V4_FLOW:
2785                         if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2786                             (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2787                                 info->data = RXH_IP_SRC | RXH_IP_DST |
2788                                              RXH_L4_B_0_1 | RXH_L4_B_2_3;
2789                         else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2790                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2791                         break;
2792                 case SCTP_V4_FLOW:
2793                 case AH_ESP_V4_FLOW:
2794                 case IPV4_FLOW:
2795                         if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2796                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2797                         break;
2798                 case TCP_V6_FLOW:
2799                         if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2800                                 info->data = RXH_IP_SRC | RXH_IP_DST |
2801                                              RXH_L4_B_0_1 | RXH_L4_B_2_3;
2802                         else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2803                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2804                         break;
2805                 case UDP_V6_FLOW:
2806                         if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2807                             (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2808                                 info->data = RXH_IP_SRC | RXH_IP_DST |
2809                                              RXH_L4_B_0_1 | RXH_L4_B_2_3;
2810                         else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2811                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2812                         break;
2813                 case SCTP_V6_FLOW:
2814                 case AH_ESP_V6_FLOW:
2815                 case IPV6_FLOW:
2816                         if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2817                                 info->data = RXH_IP_SRC | RXH_IP_DST;
2818                         break;
2819                 }
2820                 return 0;
2821         }
2822         case ETHTOOL_GRXRINGS:
2823                 info->data = pi->nqsets;
2824                 return 0;
2825         }
2826         return -EOPNOTSUPP;
2827 }
2828
2829 static const struct ethtool_ops cxgb_ethtool_ops = {
2830         .get_settings      = get_settings,
2831         .set_settings      = set_settings,
2832         .get_drvinfo       = get_drvinfo,
2833         .get_msglevel      = get_msglevel,
2834         .set_msglevel      = set_msglevel,
2835         .get_ringparam     = get_sge_param,
2836         .set_ringparam     = set_sge_param,
2837         .get_coalesce      = get_coalesce,
2838         .set_coalesce      = set_coalesce,
2839         .get_eeprom_len    = get_eeprom_len,
2840         .get_eeprom        = get_eeprom,
2841         .set_eeprom        = set_eeprom,
2842         .get_pauseparam    = get_pauseparam,
2843         .set_pauseparam    = set_pauseparam,
2844         .get_link          = ethtool_op_get_link,
2845         .get_strings       = get_strings,
2846         .set_phys_id       = identify_port,
2847         .nway_reset        = restart_autoneg,
2848         .get_sset_count    = get_sset_count,
2849         .get_ethtool_stats = get_stats,
2850         .get_regs_len      = get_regs_len,
2851         .get_regs          = get_regs,
2852         .get_wol           = get_wol,
2853         .set_wol           = set_wol,
2854         .get_rxnfc         = get_rxnfc,
2855         .get_rxfh_indir_size = get_rss_table_size,
2856         .get_rxfh_indir    = get_rss_table,
2857         .set_rxfh_indir    = set_rss_table,
2858         .flash_device      = set_flash,
2859 };
2860
2861 /*
2862  * debugfs support
2863  */
2864 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2865                         loff_t *ppos)
2866 {
2867         loff_t pos = *ppos;
2868         loff_t avail = file_inode(file)->i_size;
2869         unsigned int mem = (uintptr_t)file->private_data & 3;
2870         struct adapter *adap = file->private_data - mem;
2871
2872         if (pos < 0)
2873                 return -EINVAL;
2874         if (pos >= avail)
2875                 return 0;
2876         if (count > avail - pos)
2877                 count = avail - pos;
2878
2879         while (count) {
2880                 size_t len;
2881                 int ret, ofst;
2882                 __be32 data[16];
2883
2884                 if ((mem == MEM_MC) || (mem == MEM_MC1))
2885                         ret = t4_mc_read(adap, mem % MEM_MC, pos, data, NULL);
2886                 else
2887                         ret = t4_edc_read(adap, mem, pos, data, NULL);
2888                 if (ret)
2889                         return ret;
2890
2891                 ofst = pos % sizeof(data);
2892                 len = min(count, sizeof(data) - ofst);
2893                 if (copy_to_user(buf, (u8 *)data + ofst, len))
2894                         return -EFAULT;
2895
2896                 buf += len;
2897                 pos += len;
2898                 count -= len;
2899         }
2900         count = pos - *ppos;
2901         *ppos = pos;
2902         return count;
2903 }
2904
2905 static const struct file_operations mem_debugfs_fops = {
2906         .owner   = THIS_MODULE,
2907         .open    = simple_open,
2908         .read    = mem_read,
2909         .llseek  = default_llseek,
2910 };
2911
2912 static void add_debugfs_mem(struct adapter *adap, const char *name,
2913                             unsigned int idx, unsigned int size_mb)
2914 {
2915         struct dentry *de;
2916
2917         de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2918                                  (void *)adap + idx, &mem_debugfs_fops);
2919         if (de && de->d_inode)
2920                 de->d_inode->i_size = size_mb << 20;
2921 }
2922
2923 static int setup_debugfs(struct adapter *adap)
2924 {
2925         int i;
2926         u32 size;
2927
2928         if (IS_ERR_OR_NULL(adap->debugfs_root))
2929                 return -1;
2930
2931         i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2932         if (i & EDRAM0_ENABLE) {
2933                 size = t4_read_reg(adap, MA_EDRAM0_BAR);
2934                 add_debugfs_mem(adap, "edc0", MEM_EDC0, EDRAM_SIZE_GET(size));
2935         }
2936         if (i & EDRAM1_ENABLE) {
2937                 size = t4_read_reg(adap, MA_EDRAM1_BAR);
2938                 add_debugfs_mem(adap, "edc1", MEM_EDC1, EDRAM_SIZE_GET(size));
2939         }
2940         if (is_t4(adap->chip)) {
2941                 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2942                 if (i & EXT_MEM_ENABLE)
2943                         add_debugfs_mem(adap, "mc", MEM_MC,
2944                                         EXT_MEM_SIZE_GET(size));
2945         } else {
2946                 if (i & EXT_MEM_ENABLE) {
2947                         size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2948                         add_debugfs_mem(adap, "mc0", MEM_MC0,
2949                                         EXT_MEM_SIZE_GET(size));
2950                 }
2951                 if (i & EXT_MEM1_ENABLE) {
2952                         size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR);
2953                         add_debugfs_mem(adap, "mc1", MEM_MC1,
2954                                         EXT_MEM_SIZE_GET(size));
2955                 }
2956         }
2957         if (adap->l2t)
2958                 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2959                                     &t4_l2t_fops);
2960         return 0;
2961 }
2962
2963 /*
2964  * upper-layer driver support
2965  */
2966
2967 /*
2968  * Allocate an active-open TID and set it to the supplied value.
2969  */
2970 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2971 {
2972         int atid = -1;
2973
2974         spin_lock_bh(&t->atid_lock);
2975         if (t->afree) {
2976                 union aopen_entry *p = t->afree;
2977
2978                 atid = (p - t->atid_tab) + t->atid_base;
2979                 t->afree = p->next;
2980                 p->data = data;
2981                 t->atids_in_use++;
2982         }
2983         spin_unlock_bh(&t->atid_lock);
2984         return atid;
2985 }
2986 EXPORT_SYMBOL(cxgb4_alloc_atid);
2987
2988 /*
2989  * Release an active-open TID.
2990  */
2991 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2992 {
2993         union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
2994
2995         spin_lock_bh(&t->atid_lock);
2996         p->next = t->afree;
2997         t->afree = p;
2998         t->atids_in_use--;
2999         spin_unlock_bh(&t->atid_lock);
3000 }
3001 EXPORT_SYMBOL(cxgb4_free_atid);
3002
3003 /*
3004  * Allocate a server TID and set it to the supplied value.
3005  */
3006 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
3007 {
3008         int stid;
3009
3010         spin_lock_bh(&t->stid_lock);
3011         if (family == PF_INET) {
3012                 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
3013                 if (stid < t->nstids)
3014                         __set_bit(stid, t->stid_bmap);
3015                 else
3016                         stid = -1;
3017         } else {
3018                 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
3019                 if (stid < 0)
3020                         stid = -1;
3021         }
3022         if (stid >= 0) {
3023                 t->stid_tab[stid].data = data;
3024                 stid += t->stid_base;
3025                 t->stids_in_use++;
3026         }
3027         spin_unlock_bh(&t->stid_lock);
3028         return stid;
3029 }
3030 EXPORT_SYMBOL(cxgb4_alloc_stid);
3031
3032 /* Allocate a server filter TID and set it to the supplied value.
3033  */
3034 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
3035 {
3036         int stid;
3037
3038         spin_lock_bh(&t->stid_lock);
3039         if (family == PF_INET) {
3040                 stid = find_next_zero_bit(t->stid_bmap,
3041                                 t->nstids + t->nsftids, t->nstids);
3042                 if (stid < (t->nstids + t->nsftids))
3043                         __set_bit(stid, t->stid_bmap);
3044                 else
3045                         stid = -1;
3046         } else {
3047                 stid = -1;
3048         }
3049         if (stid >= 0) {
3050                 t->stid_tab[stid].data = data;
3051                 stid += t->stid_base;
3052                 t->stids_in_use++;
3053         }
3054         spin_unlock_bh(&t->stid_lock);
3055         return stid;
3056 }
3057 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3058
3059 /* Release a server TID.
3060  */
3061 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3062 {
3063         stid -= t->stid_base;
3064         spin_lock_bh(&t->stid_lock);
3065         if (family == PF_INET)
3066                 __clear_bit(stid, t->stid_bmap);
3067         else
3068                 bitmap_release_region(t->stid_bmap, stid, 2);
3069         t->stid_tab[stid].data = NULL;
3070         t->stids_in_use--;
3071         spin_unlock_bh(&t->stid_lock);
3072 }
3073 EXPORT_SYMBOL(cxgb4_free_stid);
3074
3075 /*
3076  * Populate a TID_RELEASE WR.  Caller must properly size the skb.
3077  */
3078 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3079                            unsigned int tid)
3080 {
3081         struct cpl_tid_release *req;
3082
3083         set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3084         req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3085         INIT_TP_WR(req, tid);
3086         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3087 }
3088
3089 /*
3090  * Queue a TID release request and if necessary schedule a work queue to
3091  * process it.
3092  */
3093 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3094                                     unsigned int tid)
3095 {
3096         void **p = &t->tid_tab[tid];
3097         struct adapter *adap = container_of(t, struct adapter, tids);
3098
3099         spin_lock_bh(&adap->tid_release_lock);
3100         *p = adap->tid_release_head;
3101         /* Low 2 bits encode the Tx channel number */
3102         adap->tid_release_head = (void **)((uintptr_t)p | chan);
3103         if (!adap->tid_release_task_busy) {
3104                 adap->tid_release_task_busy = true;
3105                 queue_work(workq, &adap->tid_release_task);
3106         }
3107         spin_unlock_bh(&adap->tid_release_lock);
3108 }
3109
3110 /*
3111  * Process the list of pending TID release requests.
3112  */
3113 static void process_tid_release_list(struct work_struct *work)
3114 {
3115         struct sk_buff *skb;
3116         struct adapter *adap;
3117
3118         adap = container_of(work, struct adapter, tid_release_task);
3119
3120         spin_lock_bh(&adap->tid_release_lock);
3121         while (adap->tid_release_head) {
3122                 void **p = adap->tid_release_head;
3123                 unsigned int chan = (uintptr_t)p & 3;
3124                 p = (void *)p - chan;
3125
3126                 adap->tid_release_head = *p;
3127                 *p = NULL;
3128                 spin_unlock_bh(&adap->tid_release_lock);
3129
3130                 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3131                                          GFP_KERNEL)))
3132                         schedule_timeout_uninterruptible(1);
3133
3134                 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3135                 t4_ofld_send(adap, skb);
3136                 spin_lock_bh(&adap->tid_release_lock);
3137         }
3138         adap->tid_release_task_busy = false;
3139         spin_unlock_bh(&adap->tid_release_lock);
3140 }
3141
3142 /*
3143  * Release a TID and inform HW.  If we are unable to allocate the release
3144  * message we defer to a work queue.
3145  */
3146 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3147 {
3148         void *old;
3149         struct sk_buff *skb;
3150         struct adapter *adap = container_of(t, struct adapter, tids);
3151
3152         old = t->tid_tab[tid];
3153         skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3154         if (likely(skb)) {
3155                 t->tid_tab[tid] = NULL;
3156                 mk_tid_release(skb, chan, tid);
3157                 t4_ofld_send(adap, skb);
3158         } else
3159                 cxgb4_queue_tid_release(t, chan, tid);
3160         if (old)
3161                 atomic_dec(&t->tids_in_use);
3162 }
3163 EXPORT_SYMBOL(cxgb4_remove_tid);
3164
3165 /*
3166  * Allocate and initialize the TID tables.  Returns 0 on success.
3167  */
3168 static int tid_init(struct tid_info *t)
3169 {
3170         size_t size;
3171         unsigned int stid_bmap_size;
3172         unsigned int natids = t->natids;
3173
3174         stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3175         size = t->ntids * sizeof(*t->tid_tab) +
3176                natids * sizeof(*t->atid_tab) +
3177                t->nstids * sizeof(*t->stid_tab) +
3178                t->nsftids * sizeof(*t->stid_tab) +
3179                stid_bmap_size * sizeof(long) +
3180                t->nftids * sizeof(*t->ftid_tab) +
3181                t->nsftids * sizeof(*t->ftid_tab);
3182
3183         t->tid_tab = t4_alloc_mem(size);
3184         if (!t->tid_tab)
3185                 return -ENOMEM;
3186
3187         t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3188         t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3189         t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3190         t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3191         spin_lock_init(&t->stid_lock);
3192         spin_lock_init(&t->atid_lock);
3193
3194         t->stids_in_use = 0;
3195         t->afree = NULL;
3196         t->atids_in_use = 0;
3197         atomic_set(&t->tids_in_use, 0);
3198
3199         /* Setup the free list for atid_tab and clear the stid bitmap. */
3200         if (natids) {
3201                 while (--natids)
3202                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3203                 t->afree = t->atid_tab;
3204         }
3205         bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3206         return 0;
3207 }
3208
3209 /**
3210  *      cxgb4_create_server - create an IP server
3211  *      @dev: the device
3212  *      @stid: the server TID
3213  *      @sip: local IP address to bind server to
3214  *      @sport: the server's TCP port
3215  *      @queue: queue to direct messages from this server to
3216  *
3217  *      Create an IP server for the given port and address.
3218  *      Returns <0 on error and one of the %NET_XMIT_* values on success.
3219  */
3220 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3221                         __be32 sip, __be16 sport, __be16 vlan,
3222                         unsigned int queue)
3223 {
3224         unsigned int chan;
3225         struct sk_buff *skb;
3226         struct adapter *adap;
3227         struct cpl_pass_open_req *req;
3228
3229         skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3230         if (!skb)
3231                 return -ENOMEM;
3232
3233         adap = netdev2adap(dev);
3234         req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3235         INIT_TP_WR(req, 0);
3236         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3237         req->local_port = sport;
3238         req->peer_port = htons(0);
3239         req->local_ip = sip;
3240         req->peer_ip = htonl(0);
3241         chan = rxq_to_chan(&adap->sge, queue);
3242         req->opt0 = cpu_to_be64(TX_CHAN(chan));
3243         req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3244                                 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3245         return t4_mgmt_tx(adap, skb);
3246 }
3247 EXPORT_SYMBOL(cxgb4_create_server);
3248
3249 /**
3250  *      cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3251  *      @mtus: the HW MTU table
3252  *      @mtu: the target MTU
3253  *      @idx: index of selected entry in the MTU table
3254  *
3255  *      Returns the index and the value in the HW MTU table that is closest to
3256  *      but does not exceed @mtu, unless @mtu is smaller than any value in the
3257  *      table, in which case that smallest available value is selected.
3258  */
3259 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3260                             unsigned int *idx)
3261 {
3262         unsigned int i = 0;
3263
3264         while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3265                 ++i;
3266         if (idx)
3267                 *idx = i;
3268         return mtus[i];
3269 }
3270 EXPORT_SYMBOL(cxgb4_best_mtu);
3271
3272 /**
3273  *      cxgb4_port_chan - get the HW channel of a port
3274  *      @dev: the net device for the port
3275  *
3276  *      Return the HW Tx channel of the given port.
3277  */
3278 unsigned int cxgb4_port_chan(const struct net_device *dev)
3279 {
3280         return netdev2pinfo(dev)->tx_chan;
3281 }
3282 EXPORT_SYMBOL(cxgb4_port_chan);
3283
3284 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3285 {
3286         struct adapter *adap = netdev2adap(dev);
3287         u32 v1, v2, lp_count, hp_count;
3288
3289         v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3290         v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3291         if (is_t4(adap->chip)) {
3292                 lp_count = G_LP_COUNT(v1);
3293                 hp_count = G_HP_COUNT(v1);
3294         } else {
3295                 lp_count = G_LP_COUNT_T5(v1);
3296                 hp_count = G_HP_COUNT_T5(v2);
3297         }
3298         return lpfifo ? lp_count : hp_count;
3299 }
3300 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3301
3302 /**
3303  *      cxgb4_port_viid - get the VI id of a port
3304  *      @dev: the net device for the port
3305  *
3306  *      Return the VI id of the given port.
3307  */
3308 unsigned int cxgb4_port_viid(const struct net_device *dev)
3309 {
3310         return netdev2pinfo(dev)->viid;
3311 }
3312 EXPORT_SYMBOL(cxgb4_port_viid);
3313
3314 /**
3315  *      cxgb4_port_idx - get the index of a port
3316  *      @dev: the net device for the port
3317  *
3318  *      Return the index of the given port.
3319  */
3320 unsigned int cxgb4_port_idx(const struct net_device *dev)
3321 {
3322         return netdev2pinfo(dev)->port_id;
3323 }
3324 EXPORT_SYMBOL(cxgb4_port_idx);
3325
3326 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3327                          struct tp_tcp_stats *v6)
3328 {
3329         struct adapter *adap = pci_get_drvdata(pdev);
3330
3331         spin_lock(&adap->stats_lock);
3332         t4_tp_get_tcp_stats(adap, v4, v6);
3333         spin_unlock(&adap->stats_lock);
3334 }
3335 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3336
3337 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3338                       const unsigned int *pgsz_order)
3339 {
3340         struct adapter *adap = netdev2adap(dev);
3341
3342         t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3343         t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3344                      HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3345                      HPZ3(pgsz_order[3]));
3346 }
3347 EXPORT_SYMBOL(cxgb4_iscsi_init);
3348
3349 int cxgb4_flush_eq_cache(struct net_device *dev)
3350 {
3351         struct adapter *adap = netdev2adap(dev);
3352         int ret;
3353
3354         ret = t4_fwaddrspace_write(adap, adap->mbox,
3355                                    0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3356         return ret;
3357 }
3358 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3359
3360 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3361 {
3362         u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3363         __be64 indices;
3364         int ret;
3365
3366         ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3367         if (!ret) {
3368                 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3369                 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3370         }
3371         return ret;
3372 }
3373
3374 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3375                         u16 size)
3376 {
3377         struct adapter *adap = netdev2adap(dev);
3378         u16 hw_pidx, hw_cidx;
3379         int ret;
3380
3381         ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3382         if (ret)
3383                 goto out;
3384
3385         if (pidx != hw_pidx) {
3386                 u16 delta;
3387
3388                 if (pidx >= hw_pidx)
3389                         delta = pidx - hw_pidx;
3390                 else
3391                         delta = size - hw_pidx + pidx;
3392                 wmb();
3393                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3394                              QID(qid) | PIDX(delta));
3395         }
3396 out:
3397         return ret;
3398 }
3399 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3400
3401 static struct pci_driver cxgb4_driver;
3402
3403 static void check_neigh_update(struct neighbour *neigh)
3404 {
3405         const struct device *parent;
3406         const struct net_device *netdev = neigh->dev;
3407
3408         if (netdev->priv_flags & IFF_802_1Q_VLAN)
3409                 netdev = vlan_dev_real_dev(netdev);
3410         parent = netdev->dev.parent;
3411         if (parent && parent->driver == &cxgb4_driver.driver)
3412                 t4_l2t_update(dev_get_drvdata(parent), neigh);
3413 }
3414
3415 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3416                        void *data)
3417 {
3418         switch (event) {
3419         case NETEVENT_NEIGH_UPDATE:
3420                 check_neigh_update(data);
3421                 break;
3422         case NETEVENT_REDIRECT:
3423         default:
3424                 break;
3425         }
3426         return 0;
3427 }
3428
3429 static bool netevent_registered;
3430 static struct notifier_block cxgb4_netevent_nb = {
3431         .notifier_call = netevent_cb
3432 };
3433
3434 static void drain_db_fifo(struct adapter *adap, int usecs)
3435 {
3436         u32 v1, v2, lp_count, hp_count;
3437
3438         do {
3439                 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3440                 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3441                 if (is_t4(adap->chip)) {
3442                         lp_count = G_LP_COUNT(v1);
3443                         hp_count = G_HP_COUNT(v1);
3444                 } else {
3445                         lp_count = G_LP_COUNT_T5(v1);
3446                         hp_count = G_HP_COUNT_T5(v2);
3447                 }
3448
3449                 if (lp_count == 0 && hp_count == 0)
3450                         break;
3451                 set_current_state(TASK_UNINTERRUPTIBLE);
3452                 schedule_timeout(usecs_to_jiffies(usecs));
3453         } while (1);
3454 }
3455
3456 static void disable_txq_db(struct sge_txq *q)
3457 {
3458         spin_lock_irq(&q->db_lock);
3459         q->db_disabled = 1;
3460         spin_unlock_irq(&q->db_lock);
3461 }
3462
3463 static void enable_txq_db(struct sge_txq *q)
3464 {
3465         spin_lock_irq(&q->db_lock);
3466         q->db_disabled = 0;
3467         spin_unlock_irq(&q->db_lock);
3468 }
3469
3470 static void disable_dbs(struct adapter *adap)
3471 {
3472         int i;
3473
3474         for_each_ethrxq(&adap->sge, i)
3475                 disable_txq_db(&adap->sge.ethtxq[i].q);
3476         for_each_ofldrxq(&adap->sge, i)
3477                 disable_txq_db(&adap->sge.ofldtxq[i].q);
3478         for_each_port(adap, i)
3479                 disable_txq_db(&adap->sge.ctrlq[i].q);
3480 }
3481
3482 static void enable_dbs(struct adapter *adap)
3483 {
3484         int i;
3485
3486         for_each_ethrxq(&adap->sge, i)
3487                 enable_txq_db(&adap->sge.ethtxq[i].q);
3488         for_each_ofldrxq(&adap->sge, i)
3489                 enable_txq_db(&adap->sge.ofldtxq[i].q);
3490         for_each_port(adap, i)
3491                 enable_txq_db(&adap->sge.ctrlq[i].q);
3492 }
3493
3494 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3495 {
3496         u16 hw_pidx, hw_cidx;
3497         int ret;
3498
3499         spin_lock_bh(&q->db_lock);
3500         ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3501         if (ret)
3502                 goto out;
3503         if (q->db_pidx != hw_pidx) {
3504                 u16 delta;
3505
3506                 if (q->db_pidx >= hw_pidx)
3507                         delta = q->db_pidx - hw_pidx;
3508                 else
3509                         delta = q->size - hw_pidx + q->db_pidx;
3510                 wmb();
3511                 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3512                              QID(q->cntxt_id) | PIDX(delta));
3513         }
3514 out:
3515         q->db_disabled = 0;
3516         spin_unlock_bh(&q->db_lock);
3517         if (ret)
3518                 CH_WARN(adap, "DB drop recovery failed.\n");
3519 }
3520 static void recover_all_queues(struct adapter *adap)
3521 {
3522         int i;
3523
3524         for_each_ethrxq(&adap->sge, i)
3525                 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3526         for_each_ofldrxq(&adap->sge, i)
3527                 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3528         for_each_port(adap, i)
3529                 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3530 }
3531
3532 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3533 {
3534         mutex_lock(&uld_mutex);
3535         if (adap->uld_handle[CXGB4_ULD_RDMA])
3536                 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3537                                 cmd);
3538         mutex_unlock(&uld_mutex);
3539 }
3540
3541 static void process_db_full(struct work_struct *work)
3542 {
3543         struct adapter *adap;
3544
3545         adap = container_of(work, struct adapter, db_full_task);
3546
3547         notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3548         drain_db_fifo(adap, dbfifo_drain_delay);
3549         t4_set_reg_field(adap, SGE_INT_ENABLE3,
3550                          DBFIFO_HP_INT | DBFIFO_LP_INT,
3551                          DBFIFO_HP_INT | DBFIFO_LP_INT);
3552         notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3553 }
3554
3555 static void process_db_drop(struct work_struct *work)
3556 {
3557         struct adapter *adap;
3558
3559         adap = container_of(work, struct adapter, db_drop_task);
3560
3561         if (is_t4(adap->chip)) {
3562                 disable_dbs(adap);
3563                 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3564                 drain_db_fifo(adap, 1);
3565                 recover_all_queues(adap);
3566                 enable_dbs(adap);
3567         } else {
3568                 u32 dropped_db = t4_read_reg(adap, 0x010ac);
3569                 u16 qid = (dropped_db >> 15) & 0x1ffff;
3570                 u16 pidx_inc = dropped_db & 0x1fff;
3571                 unsigned int s_qpp;
3572                 unsigned short udb_density;
3573                 unsigned long qpshift;
3574                 int page;
3575                 u32 udb;
3576
3577                 dev_warn(adap->pdev_dev,
3578                          "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
3579                          dropped_db, qid,
3580                          (dropped_db >> 14) & 1,
3581                          (dropped_db >> 13) & 1,
3582                          pidx_inc);
3583
3584                 drain_db_fifo(adap, 1);
3585
3586                 s_qpp = QUEUESPERPAGEPF1 * adap->fn;
3587                 udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
3588                                 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
3589                 qpshift = PAGE_SHIFT - ilog2(udb_density);
3590                 udb = qid << qpshift;
3591                 udb &= PAGE_MASK;
3592                 page = udb / PAGE_SIZE;
3593                 udb += (qid - (page * udb_density)) * 128;
3594
3595                 writel(PIDX(pidx_inc),  adap->bar2 + udb + 8);
3596
3597                 /* Re-enable BAR2 WC */
3598                 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
3599         }
3600
3601         t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3602 }
3603
3604 void t4_db_full(struct adapter *adap)
3605 {
3606         if (is_t4(adap->chip)) {
3607                 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3608                                  DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3609                 queue_work(workq, &adap->db_full_task);
3610         }
3611 }
3612
3613 void t4_db_dropped(struct adapter *adap)
3614 {
3615         if (is_t4(adap->chip))
3616                 queue_work(workq, &adap->db_drop_task);
3617 }
3618
3619 static void uld_attach(struct adapter *adap, unsigned int uld)
3620 {
3621         void *handle;
3622         struct cxgb4_lld_info lli;
3623         unsigned short i;
3624
3625         lli.pdev = adap->pdev;
3626         lli.l2t = adap->l2t;
3627         lli.tids = &adap->tids;
3628         lli.ports = adap->port;
3629         lli.vr = &adap->vres;
3630         lli.mtus = adap->params.mtus;
3631         if (uld == CXGB4_ULD_RDMA) {
3632                 lli.rxq_ids = adap->sge.rdma_rxq;
3633                 lli.nrxq = adap->sge.rdmaqs;
3634         } else if (uld == CXGB4_ULD_ISCSI) {
3635                 lli.rxq_ids = adap->sge.ofld_rxq;
3636                 lli.nrxq = adap->sge.ofldqsets;
3637         }
3638         lli.ntxq = adap->sge.ofldqsets;
3639         lli.nchan = adap->params.nports;
3640         lli.nports = adap->params.nports;
3641         lli.wr_cred = adap->params.ofldq_wr_cred;
3642         lli.adapter_type = adap->params.rev;
3643         lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3644         lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3645                         t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3646                         (adap->fn * 4));
3647         lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3648                         t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3649                         (adap->fn * 4));
3650         lli.filt_mode = adap->filter_mode;
3651         /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3652         for (i = 0; i < NCHAN; i++)
3653                 lli.tx_modq[i] = i;
3654         lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3655         lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3656         lli.fw_vers = adap->params.fw_vers;
3657         lli.dbfifo_int_thresh = dbfifo_int_thresh;
3658         lli.sge_pktshift = adap->sge.pktshift;
3659         lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3660
3661         handle = ulds[uld].add(&lli);
3662         if (IS_ERR(handle)) {
3663                 dev_warn(adap->pdev_dev,
3664                          "could not attach to the %s driver, error %ld\n",
3665                          uld_str[uld], PTR_ERR(handle));
3666                 return;
3667         }
3668
3669         adap->uld_handle[uld] = handle;
3670
3671         if (!netevent_registered) {
3672                 register_netevent_notifier(&cxgb4_netevent_nb);
3673                 netevent_registered = true;
3674         }
3675
3676         if (adap->flags & FULL_INIT_DONE)
3677                 ulds[uld].state_change(handle, CXGB4_STATE_UP);
3678 }
3679
3680 static void attach_ulds(struct adapter *adap)
3681 {
3682         unsigned int i;
3683
3684         mutex_lock(&uld_mutex);
3685         list_add_tail(&adap->list_node, &adapter_list);
3686         for (i = 0; i < CXGB4_ULD_MAX; i++)
3687                 if (ulds[i].add)
3688                         uld_attach(adap, i);
3689         mutex_unlock(&uld_mutex);
3690 }
3691
3692 static void detach_ulds(struct adapter *adap)
3693 {
3694         unsigned int i;
3695
3696         mutex_lock(&uld_mutex);
3697         list_del(&adap->list_node);
3698         for (i = 0; i < CXGB4_ULD_MAX; i++)
3699                 if (adap->uld_handle[i]) {
3700                         ulds[i].state_change(adap->uld_handle[i],
3701                                              CXGB4_STATE_DETACH);
3702                         adap->uld_handle[i] = NULL;
3703                 }
3704         if (netevent_registered && list_empty(&adapter_list)) {
3705                 unregister_netevent_notifier(&cxgb4_netevent_nb);
3706                 netevent_registered = false;
3707         }
3708         mutex_unlock(&uld_mutex);
3709 }
3710
3711 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3712 {
3713         unsigned int i;
3714
3715         mutex_lock(&uld_mutex);
3716         for (i = 0; i < CXGB4_ULD_MAX; i++)
3717                 if (adap->uld_handle[i])
3718                         ulds[i].state_change(adap->uld_handle[i], new_state);
3719         mutex_unlock(&uld_mutex);
3720 }
3721
3722 /**
3723  *      cxgb4_register_uld - register an upper-layer driver
3724  *      @type: the ULD type
3725  *      @p: the ULD methods
3726  *
3727  *      Registers an upper-layer driver with this driver and notifies the ULD
3728  *      about any presently available devices that support its type.  Returns
3729  *      %-EBUSY if a ULD of the same type is already registered.
3730  */
3731 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3732 {
3733         int ret = 0;
3734         struct adapter *adap;
3735
3736         if (type >= CXGB4_ULD_MAX)
3737                 return -EINVAL;
3738         mutex_lock(&uld_mutex);
3739         if (ulds[type].add) {
3740                 ret = -EBUSY;
3741                 goto out;
3742         }
3743         ulds[type] = *p;
3744         list_for_each_entry(adap, &adapter_list, list_node)
3745                 uld_attach(adap, type);
3746 out:    mutex_unlock(&uld_mutex);
3747         return ret;
3748 }
3749 EXPORT_SYMBOL(cxgb4_register_uld);
3750
3751 /**
3752  *      cxgb4_unregister_uld - unregister an upper-layer driver
3753  *      @type: the ULD type
3754  *
3755  *      Unregisters an existing upper-layer driver.
3756  */
3757 int cxgb4_unregister_uld(enum cxgb4_uld type)
3758 {
3759         struct adapter *adap;
3760
3761         if (type >= CXGB4_ULD_MAX)
3762                 return -EINVAL;
3763         mutex_lock(&uld_mutex);
3764         list_for_each_entry(adap, &adapter_list, list_node)
3765                 adap->uld_handle[type] = NULL;
3766         ulds[type].add = NULL;
3767         mutex_unlock(&uld_mutex);
3768         return 0;
3769 }
3770 EXPORT_SYMBOL(cxgb4_unregister_uld);
3771
3772 /**
3773  *      cxgb_up - enable the adapter
3774  *      @adap: adapter being enabled
3775  *
3776  *      Called when the first port is enabled, this function performs the
3777  *      actions necessary to make an adapter operational, such as completing
3778  *      the initialization of HW modules, and enabling interrupts.
3779  *
3780  *      Must be called with the rtnl lock held.
3781  */
3782 static int cxgb_up(struct adapter *adap)
3783 {
3784         int err;
3785
3786         err = setup_sge_queues(adap);
3787         if (err)
3788                 goto out;
3789         err = setup_rss(adap);
3790         if (err)
3791                 goto freeq;
3792
3793         if (adap->flags & USING_MSIX) {
3794                 name_msix_vecs(adap);
3795                 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
3796                                   adap->msix_info[0].desc, adap);
3797                 if (err)
3798                         goto irq_err;
3799
3800                 err = request_msix_queue_irqs(adap);
3801                 if (err) {
3802                         free_irq(adap->msix_info[0].vec, adap);
3803                         goto irq_err;
3804                 }
3805         } else {
3806                 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
3807                                   (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
3808                                   adap->port[0]->name, adap);
3809                 if (err)
3810                         goto irq_err;
3811         }
3812         enable_rx(adap);
3813         t4_sge_start(adap);
3814         t4_intr_enable(adap);
3815         adap->flags |= FULL_INIT_DONE;
3816         notify_ulds(adap, CXGB4_STATE_UP);
3817  out:
3818         return err;
3819  irq_err:
3820         dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
3821  freeq:
3822         t4_free_sge_resources(adap);
3823         goto out;
3824 }
3825
3826 static void cxgb_down(struct adapter *adapter)
3827 {
3828         t4_intr_disable(adapter);
3829         cancel_work_sync(&adapter->tid_release_task);
3830         cancel_work_sync(&adapter->db_full_task);
3831         cancel_work_sync(&adapter->db_drop_task);
3832         adapter->tid_release_task_busy = false;
3833         adapter->tid_release_head = NULL;
3834
3835         if (adapter->flags & USING_MSIX) {
3836                 free_msix_queue_irqs(adapter);
3837                 free_irq(adapter->msix_info[0].vec, adapter);
3838         } else
3839                 free_irq(adapter->pdev->irq, adapter);
3840         quiesce_rx(adapter);
3841         t4_sge_stop(adapter);
3842         t4_free_sge_resources(adapter);
3843         adapter->flags &= ~FULL_INIT_DONE;
3844 }
3845
3846 /*
3847  * net_device operations
3848  */
3849 static int cxgb_open(struct net_device *dev)
3850 {
3851         int err;
3852         struct port_info *pi = netdev_priv(dev);
3853         struct adapter *adapter = pi->adapter;
3854
3855         netif_carrier_off(dev);
3856
3857         if (!(adapter->flags & FULL_INIT_DONE)) {
3858                 err = cxgb_up(adapter);
3859                 if (err < 0)
3860                         return err;
3861         }
3862
3863         err = link_start(dev);
3864         if (!err)
3865                 netif_tx_start_all_queues(dev);
3866         return err;
3867 }
3868
3869 static int cxgb_close(struct net_device *dev)
3870 {
3871         struct port_info *pi = netdev_priv(dev);
3872         struct adapter *adapter = pi->adapter;
3873
3874         netif_tx_stop_all_queues(dev);
3875         netif_carrier_off(dev);
3876         return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
3877 }
3878
3879 /* Return an error number if the indicated filter isn't writable ...
3880  */
3881 static int writable_filter(struct filter_entry *f)
3882 {
3883         if (f->locked)
3884                 return -EPERM;
3885         if (f->pending)
3886                 return -EBUSY;
3887
3888         return 0;
3889 }
3890
3891 /* Delete the filter at the specified index (if valid).  The checks for all
3892  * the common problems with doing this like the filter being locked, currently
3893  * pending in another operation, etc.
3894  */
3895 static int delete_filter(struct adapter *adapter, unsigned int fidx)
3896 {
3897         struct filter_entry *f;
3898         int ret;
3899
3900         if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
3901                 return -EINVAL;
3902
3903         f = &adapter->tids.ftid_tab[fidx];
3904         ret = writable_filter(f);
3905         if (ret)
3906                 return ret;
3907         if (f->valid)
3908                 return del_filter_wr(adapter, fidx);
3909
3910         return 0;
3911 }
3912
3913 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
3914                 __be32 sip, __be16 sport, __be16 vlan,
3915                 unsigned int queue, unsigned char port, unsigned char mask)
3916 {
3917         int ret;
3918         struct filter_entry *f;
3919         struct adapter *adap;
3920         int i;
3921         u8 *val;
3922
3923         adap = netdev2adap(dev);
3924
3925         /* Adjust stid to correct filter index */
3926         stid -= adap->tids.nstids;
3927         stid += adap->tids.nftids;
3928
3929         /* Check to make sure the filter requested is writable ...
3930          */
3931         f = &adap->tids.ftid_tab[stid];
3932         ret = writable_filter(f);
3933         if (ret)
3934                 return ret;
3935
3936         /* Clear out any old resources being used by the filter before
3937          * we start constructing the new filter.
3938          */
3939         if (f->valid)
3940                 clear_filter(adap, f);
3941
3942         /* Clear out filter specifications */
3943         memset(&f->fs, 0, sizeof(struct ch_filter_specification));
3944         f->fs.val.lport = cpu_to_be16(sport);
3945         f->fs.mask.lport  = ~0;
3946         val = (u8 *)&sip;
3947         if ((val[0] | val[1] | val[2] | val[3]) != 0) {
3948                 for (i = 0; i < 4; i++) {
3949                         f->fs.val.lip[i] = val[i];
3950                         f->fs.mask.lip[i] = ~0;
3951                 }
3952                 if (adap->filter_mode & F_PORT) {
3953                         f->fs.val.iport = port;
3954                         f->fs.mask.iport = mask;
3955                 }
3956         }
3957
3958         f->fs.dirsteer = 1;
3959         f->fs.iq = queue;
3960         /* Mark filter as locked */
3961         f->locked = 1;
3962         f->fs.rpttid = 1;
3963
3964         ret = set_filter_wr(adap, stid);
3965         if (ret) {
3966                 clear_filter(adap, f);
3967                 return ret;
3968         }
3969
3970         return 0;
3971 }
3972 EXPORT_SYMBOL(cxgb4_create_server_filter);
3973
3974 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
3975                 unsigned int queue, bool ipv6)
3976 {
3977         int ret;
3978         struct filter_entry *f;
3979         struct adapter *adap;
3980
3981         adap = netdev2adap(dev);
3982
3983         /* Adjust stid to correct filter index */
3984         stid -= adap->tids.nstids;
3985         stid += adap->tids.nftids;
3986
3987         f = &adap->tids.ftid_tab[stid];
3988         /* Unlock the filter */
3989         f->locked = 0;
3990
3991         ret = delete_filter(adap, stid);
3992         if (ret)
3993                 return ret;
3994
3995         return 0;
3996 }
3997 EXPORT_SYMBOL(cxgb4_remove_server_filter);
3998
3999 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
4000                                                 struct rtnl_link_stats64 *ns)
4001 {
4002         struct port_stats stats;
4003         struct port_info *p = netdev_priv(dev);
4004         struct adapter *adapter = p->adapter;
4005
4006         spin_lock(&adapter->stats_lock);
4007         t4_get_port_stats(adapter, p->tx_chan, &stats);
4008         spin_unlock(&adapter->stats_lock);
4009
4010         ns->tx_bytes   = stats.tx_octets;
4011         ns->tx_packets = stats.tx_frames;
4012         ns->rx_bytes   = stats.rx_octets;
4013         ns->rx_packets = stats.rx_frames;
4014         ns->multicast  = stats.rx_mcast_frames;
4015
4016         /* detailed rx_errors */
4017         ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
4018                                stats.rx_runt;
4019         ns->rx_over_errors   = 0;
4020         ns->rx_crc_errors    = stats.rx_fcs_err;
4021         ns->rx_frame_errors  = stats.rx_symbol_err;
4022         ns->rx_fifo_errors   = stats.rx_ovflow0 + stats.rx_ovflow1 +
4023                                stats.rx_ovflow2 + stats.rx_ovflow3 +
4024                                stats.rx_trunc0 + stats.rx_trunc1 +
4025                                stats.rx_trunc2 + stats.rx_trunc3;
4026         ns->rx_missed_errors = 0;
4027
4028         /* detailed tx_errors */
4029         ns->tx_aborted_errors   = 0;
4030         ns->tx_carrier_errors   = 0;
4031         ns->tx_fifo_errors      = 0;
4032         ns->tx_heartbeat_errors = 0;
4033         ns->tx_window_errors    = 0;
4034
4035         ns->tx_errors = stats.tx_error_frames;
4036         ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
4037                 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
4038         return ns;
4039 }
4040
4041 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
4042 {
4043         unsigned int mbox;
4044         int ret = 0, prtad, devad;
4045         struct port_info *pi = netdev_priv(dev);
4046         struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
4047
4048         switch (cmd) {
4049         case SIOCGMIIPHY:
4050                 if (pi->mdio_addr < 0)
4051                         return -EOPNOTSUPP;
4052                 data->phy_id = pi->mdio_addr;
4053                 break;
4054         case SIOCGMIIREG:
4055         case SIOCSMIIREG:
4056                 if (mdio_phy_id_is_c45(data->phy_id)) {
4057                         prtad = mdio_phy_id_prtad(data->phy_id);
4058                         devad = mdio_phy_id_devad(data->phy_id);
4059                 } else if (data->phy_id < 32) {
4060                         prtad = data->phy_id;
4061                         devad = 0;
4062                         data->reg_num &= 0x1f;
4063                 } else
4064                         return -EINVAL;
4065
4066                 mbox = pi->adapter->fn;
4067                 if (cmd == SIOCGMIIREG)
4068                         ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
4069                                          data->reg_num, &data->val_out);
4070                 else
4071                         ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
4072                                          data->reg_num, data->val_in);
4073                 break;
4074         default:
4075                 return -EOPNOTSUPP;
4076         }
4077         return ret;
4078 }
4079
4080 static void cxgb_set_rxmode(struct net_device *dev)
4081 {
4082         /* unfortunately we can't return errors to the stack */
4083         set_rxmode(dev, -1, false);
4084 }
4085
4086 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
4087 {
4088         int ret;
4089         struct port_info *pi = netdev_priv(dev);
4090
4091         if (new_mtu < 81 || new_mtu > MAX_MTU)         /* accommodate SACK */
4092                 return -EINVAL;
4093         ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
4094                             -1, -1, -1, true);
4095         if (!ret)
4096                 dev->mtu = new_mtu;
4097         return ret;
4098 }
4099
4100 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
4101 {
4102         int ret;
4103         struct sockaddr *addr = p;
4104         struct port_info *pi = netdev_priv(dev);
4105
4106         if (!is_valid_ether_addr(addr->sa_data))
4107                 return -EADDRNOTAVAIL;
4108
4109         ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
4110                             pi->xact_addr_filt, addr->sa_data, true, true);
4111         if (ret < 0)
4112                 return ret;
4113
4114         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4115         pi->xact_addr_filt = ret;
4116         return 0;
4117 }
4118
4119 #ifdef CONFIG_NET_POLL_CONTROLLER
4120 static void cxgb_netpoll(struct net_device *dev)
4121 {
4122         struct port_info *pi = netdev_priv(dev);
4123         struct adapter *adap = pi->adapter;
4124
4125         if (adap->flags & USING_MSIX) {
4126                 int i;
4127                 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4128
4129                 for (i = pi->nqsets; i; i--, rx++)
4130                         t4_sge_intr_msix(0, &rx->rspq);
4131         } else
4132                 t4_intr_handler(adap)(0, adap);
4133 }
4134 #endif
4135
4136 static const struct net_device_ops cxgb4_netdev_ops = {
4137         .ndo_open             = cxgb_open,
4138         .ndo_stop             = cxgb_close,
4139         .ndo_start_xmit       = t4_eth_xmit,
4140         .ndo_get_stats64      = cxgb_get_stats,
4141         .ndo_set_rx_mode      = cxgb_set_rxmode,
4142         .ndo_set_mac_address  = cxgb_set_mac_addr,
4143         .ndo_set_features     = cxgb_set_features,
4144         .ndo_validate_addr    = eth_validate_addr,
4145         .ndo_do_ioctl         = cxgb_ioctl,
4146         .ndo_change_mtu       = cxgb_change_mtu,
4147 #ifdef CONFIG_NET_POLL_CONTROLLER
4148         .ndo_poll_controller  = cxgb_netpoll,
4149 #endif
4150 };
4151
4152 void t4_fatal_err(struct adapter *adap)
4153 {
4154         t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4155         t4_intr_disable(adap);
4156         dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4157 }
4158
4159 static void setup_memwin(struct adapter *adap)
4160 {
4161         u32 bar0, mem_win0_base, mem_win1_base, mem_win2_base;
4162
4163         bar0 = pci_resource_start(adap->pdev, 0);  /* truncation intentional */
4164         if (is_t4(adap->chip)) {
4165                 mem_win0_base = bar0 + MEMWIN0_BASE;
4166                 mem_win1_base = bar0 + MEMWIN1_BASE;
4167                 mem_win2_base = bar0 + MEMWIN2_BASE;
4168         } else {
4169                 /* For T5, only relative offset inside the PCIe BAR is passed */
4170                 mem_win0_base = MEMWIN0_BASE;
4171                 mem_win1_base = MEMWIN1_BASE_T5;
4172                 mem_win2_base = MEMWIN2_BASE_T5;
4173         }
4174         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4175                      mem_win0_base | BIR(0) |
4176                      WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4177         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4178                      mem_win1_base | BIR(0) |
4179                      WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4180         t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4181                      mem_win2_base | BIR(0) |
4182                      WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
4183 }
4184
4185 static void setup_memwin_rdma(struct adapter *adap)
4186 {
4187         if (adap->vres.ocq.size) {
4188                 unsigned int start, sz_kb;
4189
4190                 start = pci_resource_start(adap->pdev, 2) +
4191                         OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4192                 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4193                 t4_write_reg(adap,
4194                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4195                              start | BIR(1) | WINDOW(ilog2(sz_kb)));
4196                 t4_write_reg(adap,
4197                              PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4198                              adap->vres.ocq.start);
4199                 t4_read_reg(adap,
4200                             PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4201         }
4202 }
4203
4204 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4205 {
4206         u32 v;
4207         int ret;
4208
4209         /* get device capabilities */
4210         memset(c, 0, sizeof(*c));
4211         c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4212                                FW_CMD_REQUEST | FW_CMD_READ);
4213         c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4214         ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4215         if (ret < 0)
4216                 return ret;
4217
4218         /* select capabilities we'll be using */
4219         if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4220                 if (!vf_acls)
4221                         c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4222                 else
4223                         c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4224         } else if (vf_acls) {
4225                 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4226                 return ret;
4227         }
4228         c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4229                                FW_CMD_REQUEST | FW_CMD_WRITE);
4230         ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4231         if (ret < 0)
4232                 return ret;
4233
4234         ret = t4_config_glbl_rss(adap, adap->fn,
4235                                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4236                                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4237                                  FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4238         if (ret < 0)
4239                 return ret;
4240
4241         ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4242                           0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4243         if (ret < 0)
4244                 return ret;
4245
4246         t4_sge_init(adap);
4247
4248         /* tweak some settings */
4249         t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4250         t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4251         t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4252         v = t4_read_reg(adap, TP_PIO_DATA);
4253         t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4254
4255         /* first 4 Tx modulation queues point to consecutive Tx channels */
4256         adap->params.tp.tx_modq_map = 0xE4;
4257         t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4258                      V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4259
4260         /* associate each Tx modulation queue with consecutive Tx channels */
4261         v = 0x84218421;
4262         t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4263                           &v, 1, A_TP_TX_SCHED_HDR);
4264         t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4265                           &v, 1, A_TP_TX_SCHED_FIFO);
4266         t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4267                           &v, 1, A_TP_TX_SCHED_PCMD);
4268
4269 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4270         if (is_offload(adap)) {
4271                 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4272                              V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4273                              V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4274                              V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4275                              V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4276                 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4277                              V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4278                              V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4279                              V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4280                              V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4281         }
4282
4283         /* get basic stuff going */
4284         return t4_early_init(adap, adap->fn);
4285 }
4286
4287 /*
4288  * Max # of ATIDs.  The absolute HW max is 16K but we keep it lower.
4289  */
4290 #define MAX_ATIDS 8192U
4291
4292 /*
4293  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4294  *
4295  * If the firmware we're dealing with has Configuration File support, then
4296  * we use that to perform all configuration
4297  */
4298
4299 /*
4300  * Tweak configuration based on module parameters, etc.  Most of these have
4301  * defaults assigned to them by Firmware Configuration Files (if we're using
4302  * them) but need to be explicitly set if we're using hard-coded
4303  * initialization.  But even in the case of using Firmware Configuration
4304  * Files, we'd like to expose the ability to change these via module
4305  * parameters so these are essentially common tweaks/settings for
4306  * Configuration Files and hard-coded initialization ...
4307  */
4308 static int adap_init0_tweaks(struct adapter *adapter)
4309 {
4310         /*
4311          * Fix up various Host-Dependent Parameters like Page Size, Cache
4312          * Line Size, etc.  The firmware default is for a 4KB Page Size and
4313          * 64B Cache Line Size ...
4314          */
4315         t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4316
4317         /*
4318          * Process module parameters which affect early initialization.
4319          */
4320         if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4321                 dev_err(&adapter->pdev->dev,
4322                         "Ignoring illegal rx_dma_offset=%d, using 2\n",
4323                         rx_dma_offset);
4324                 rx_dma_offset = 2;
4325         }
4326         t4_set_reg_field(adapter, SGE_CONTROL,
4327                          PKTSHIFT_MASK,
4328                          PKTSHIFT(rx_dma_offset));
4329
4330         /*
4331          * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4332          * adds the pseudo header itself.
4333          */
4334         t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
4335                                CSUM_HAS_PSEUDO_HDR, 0);
4336
4337         return 0;
4338 }
4339
4340 /*
4341  * Attempt to initialize the adapter via a Firmware Configuration File.
4342  */
4343 static int adap_init0_config(struct adapter *adapter, int reset)
4344 {
4345         struct fw_caps_config_cmd caps_cmd;
4346         const struct firmware *cf;
4347         unsigned long mtype = 0, maddr = 0;
4348         u32 finiver, finicsum, cfcsum;
4349         int ret, using_flash;
4350         char *fw_config_file, fw_config_file_path[256];
4351
4352         /*
4353          * Reset device if necessary.
4354          */
4355         if (reset) {
4356                 ret = t4_fw_reset(adapter, adapter->mbox,
4357                                   PIORSTMODE | PIORST);
4358                 if (ret < 0)
4359                         goto bye;
4360         }
4361
4362         /*
4363          * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4364          * then use that.  Otherwise, use the configuration file stored
4365          * in the adapter flash ...
4366          */
4367         switch (CHELSIO_CHIP_VERSION(adapter->chip)) {
4368         case CHELSIO_T4:
4369                 fw_config_file = FW_CFNAME;
4370                 break;
4371         case CHELSIO_T5:
4372                 fw_config_file = FW5_CFNAME;
4373                 break;
4374         default:
4375                 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4376                        adapter->pdev->device);
4377                 ret = -EINVAL;
4378                 goto bye;
4379         }
4380
4381         ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4382         if (ret < 0) {
4383                 using_flash = 1;
4384                 mtype = FW_MEMTYPE_CF_FLASH;
4385                 maddr = t4_flash_cfg_addr(adapter);
4386         } else {
4387                 u32 params[7], val[7];
4388
4389                 using_flash = 0;
4390                 if (cf->size >= FLASH_CFG_MAX_SIZE)
4391                         ret = -ENOMEM;
4392                 else {
4393                         params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4394                              FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4395                         ret = t4_query_params(adapter, adapter->mbox,
4396                                               adapter->fn, 0, 1, params, val);
4397                         if (ret == 0) {
4398                                 /*
4399                                  * For t4_memory_write() below addresses and
4400                                  * sizes have to be in terms of multiples of 4
4401                                  * bytes.  So, if the Configuration File isn't
4402                                  * a multiple of 4 bytes in length we'll have
4403                                  * to write that out separately since we can't
4404                                  * guarantee that the bytes following the
4405                                  * residual byte in the buffer returned by
4406                                  * request_firmware() are zeroed out ...
4407                                  */
4408                                 size_t resid = cf->size & 0x3;
4409                                 size_t size = cf->size & ~0x3;
4410                                 __be32 *data = (__be32 *)cf->data;
4411
4412                                 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
4413                                 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
4414
4415                                 ret = t4_memory_write(adapter, mtype, maddr,
4416                                                       size, data);
4417                                 if (ret == 0 && resid != 0) {
4418                                         union {
4419                                                 __be32 word;
4420                                                 char buf[4];
4421                                         } last;
4422                                         int i;
4423
4424                                         last.word = data[size >> 2];
4425                                         for (i = resid; i < 4; i++)
4426                                                 last.buf[i] = 0;
4427                                         ret = t4_memory_write(adapter, mtype,
4428                                                               maddr + size,
4429                                                               4, &last.word);
4430                                 }
4431                         }
4432                 }
4433
4434                 release_firmware(cf);
4435                 if (ret)
4436                         goto bye;
4437         }
4438
4439         /*
4440          * Issue a Capability Configuration command to the firmware to get it
4441          * to parse the Configuration File.  We don't use t4_fw_config_file()
4442          * because we want the ability to modify various features after we've
4443          * processed the configuration file ...
4444          */
4445         memset(&caps_cmd, 0, sizeof(caps_cmd));
4446         caps_cmd.op_to_write =
4447                 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4448                       FW_CMD_REQUEST |
4449                       FW_CMD_READ);
4450         caps_cmd.cfvalid_to_len16 =
4451                 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
4452                       FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4453                       FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
4454                       FW_LEN16(caps_cmd));
4455         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4456                          &caps_cmd);
4457         if (ret < 0)
4458                 goto bye;
4459
4460         finiver = ntohl(caps_cmd.finiver);
4461         finicsum = ntohl(caps_cmd.finicsum);
4462         cfcsum = ntohl(caps_cmd.cfcsum);
4463         if (finicsum != cfcsum)
4464                 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4465                          "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4466                          finicsum, cfcsum);
4467
4468         /*
4469          * And now tell the firmware to use the configuration we just loaded.
4470          */
4471         caps_cmd.op_to_write =
4472                 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4473                       FW_CMD_REQUEST |
4474                       FW_CMD_WRITE);
4475         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4476         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4477                          NULL);
4478         if (ret < 0)
4479                 goto bye;
4480
4481         /*
4482          * Tweak configuration based on system architecture, module
4483          * parameters, etc.
4484          */
4485         ret = adap_init0_tweaks(adapter);
4486         if (ret < 0)
4487                 goto bye;
4488
4489         /*
4490          * And finally tell the firmware to initialize itself using the
4491          * parameters from the Configuration File.
4492          */
4493         ret = t4_fw_initialize(adapter, adapter->mbox);
4494         if (ret < 0)
4495                 goto bye;
4496
4497         sprintf(fw_config_file_path, "/lib/firmware/%s", fw_config_file);
4498         /*
4499          * Return successfully and note that we're operating with parameters
4500          * not supplied by the driver, rather than from hard-wired
4501          * initialization constants burried in the driver.
4502          */
4503         adapter->flags |= USING_SOFT_PARAMS;
4504         dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4505                  "Configuration File %s, version %#x, computed checksum %#x\n",
4506                  (using_flash
4507                   ? "in device FLASH"
4508                   : fw_config_file_path),
4509                  finiver, cfcsum);
4510         return 0;
4511
4512         /*
4513          * Something bad happened.  Return the error ...  (If the "error"
4514          * is that there's no Configuration File on the adapter we don't
4515          * want to issue a warning since this is fairly common.)
4516          */
4517 bye:
4518         if (ret != -ENOENT)
4519                 dev_warn(adapter->pdev_dev, "Configuration file error %d\n",
4520                          -ret);
4521         return ret;
4522 }
4523
4524 /*
4525  * Attempt to initialize the adapter via hard-coded, driver supplied
4526  * parameters ...
4527  */
4528 static int adap_init0_no_config(struct adapter *adapter, int reset)
4529 {
4530         struct sge *s = &adapter->sge;
4531         struct fw_caps_config_cmd caps_cmd;
4532         u32 v;
4533         int i, ret;
4534
4535         /*
4536          * Reset device if necessary
4537          */
4538         if (reset) {
4539                 ret = t4_fw_reset(adapter, adapter->mbox,
4540                                   PIORSTMODE | PIORST);
4541                 if (ret < 0)
4542                         goto bye;
4543         }
4544
4545         /*
4546          * Get device capabilities and select which we'll be using.
4547          */
4548         memset(&caps_cmd, 0, sizeof(caps_cmd));
4549         caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4550                                      FW_CMD_REQUEST | FW_CMD_READ);
4551         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4552         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4553                          &caps_cmd);
4554         if (ret < 0)
4555                 goto bye;
4556
4557         if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4558                 if (!vf_acls)
4559                         caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4560                 else
4561                         caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4562         } else if (vf_acls) {
4563                 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
4564                 goto bye;
4565         }
4566         caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4567                               FW_CMD_REQUEST | FW_CMD_WRITE);
4568         ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4569                          NULL);
4570         if (ret < 0)
4571                 goto bye;
4572
4573         /*
4574          * Tweak configuration based on system architecture, module
4575          * parameters, etc.
4576          */
4577         ret = adap_init0_tweaks(adapter);
4578         if (ret < 0)
4579                 goto bye;
4580
4581         /*
4582          * Select RSS Global Mode we want to use.  We use "Basic Virtual"
4583          * mode which maps each Virtual Interface to its own section of
4584          * the RSS Table and we turn on all map and hash enables ...
4585          */
4586         adapter->flags |= RSS_TNLALLLOOKUP;
4587         ret = t4_config_glbl_rss(adapter, adapter->mbox,
4588                                  FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4589                                  FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4590                                  FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4591                                  ((adapter->flags & RSS_TNLALLLOOKUP) ?
4592                                         FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4593         if (ret < 0)
4594                 goto bye;
4595
4596         /*
4597          * Set up our own fundamental resource provisioning ...
4598          */
4599         ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4600                           PFRES_NEQ, PFRES_NETHCTRL,
4601                           PFRES_NIQFLINT, PFRES_NIQ,
4602                           PFRES_TC, PFRES_NVI,
4603                           FW_PFVF_CMD_CMASK_MASK,
4604                           pfvfres_pmask(adapter, adapter->fn, 0),
4605                           PFRES_NEXACTF,
4606                           PFRES_R_CAPS, PFRES_WX_CAPS);
4607         if (ret < 0)
4608                 goto bye;
4609
4610         /*
4611          * Perform low level SGE initialization.  We need to do this before we
4612          * send the firmware the INITIALIZE command because that will cause
4613          * any other PF Drivers which are waiting for the Master
4614          * Initialization to proceed forward.
4615          */
4616         for (i = 0; i < SGE_NTIMERS - 1; i++)
4617                 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4618         s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4619         s->counter_val[0] = 1;
4620         for (i = 1; i < SGE_NCOUNTERS; i++)
4621                 s->counter_val[i] = min(intr_cnt[i - 1],
4622                                         THRESHOLD_0_GET(THRESHOLD_0_MASK));
4623         t4_sge_init(adapter);
4624
4625 #ifdef CONFIG_PCI_IOV
4626         /*
4627          * Provision resource limits for Virtual Functions.  We currently
4628          * grant them all the same static resource limits except for the Port
4629          * Access Rights Mask which we're assigning based on the PF.  All of
4630          * the static provisioning stuff for both the PF and VF really needs
4631          * to be managed in a persistent manner for each device which the
4632          * firmware controls.
4633          */
4634         {
4635                 int pf, vf;
4636                 int max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
4637                                 NUM_OF_PF_WITH_SRIOV_T5;
4638
4639                 for (pf = 0; pf < max_no_pf; pf++) {
4640                         if (num_vf[pf] <= 0)
4641                                 continue;
4642
4643                         /* VF numbering starts at 1! */
4644                         for (vf = 1; vf <= num_vf[pf]; vf++) {
4645                                 ret = t4_cfg_pfvf(adapter, adapter->mbox,
4646                                                   pf, vf,
4647                                                   VFRES_NEQ, VFRES_NETHCTRL,
4648                                                   VFRES_NIQFLINT, VFRES_NIQ,
4649                                                   VFRES_TC, VFRES_NVI,
4650                                                   FW_PFVF_CMD_CMASK_MASK,
4651                                                   pfvfres_pmask(
4652                                                   adapter, pf, vf),
4653                                                   VFRES_NEXACTF,
4654                                                   VFRES_R_CAPS, VFRES_WX_CAPS);
4655                                 if (ret < 0)
4656                                         dev_warn(adapter->pdev_dev,
4657                                                  "failed to "\
4658                                                  "provision pf/vf=%d/%d; "
4659                                                  "err=%d\n", pf, vf, ret);
4660                         }
4661                 }
4662         }
4663 #endif
4664
4665         /*
4666          * Set up the default filter mode.  Later we'll want to implement this
4667          * via a firmware command, etc. ...  This needs to be done before the
4668          * firmare initialization command ...  If the selected set of fields
4669          * isn't equal to the default value, we'll need to make sure that the
4670          * field selections will fit in the 36-bit budget.
4671          */
4672         if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
4673                 int j, bits = 0;
4674
4675                 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
4676                         switch (tp_vlan_pri_map & (1 << j)) {
4677                         case 0:
4678                                 /* compressed filter field not enabled */
4679                                 break;
4680                         case FCOE_MASK:
4681                                 bits +=  1;
4682                                 break;
4683                         case PORT_MASK:
4684                                 bits +=  3;
4685                                 break;
4686                         case VNIC_ID_MASK:
4687                                 bits += 17;
4688                                 break;
4689                         case VLAN_MASK:
4690                                 bits += 17;
4691                                 break;
4692                         case TOS_MASK:
4693                                 bits +=  8;
4694                                 break;
4695                         case PROTOCOL_MASK:
4696                                 bits +=  8;
4697                                 break;
4698                         case ETHERTYPE_MASK:
4699                                 bits += 16;
4700                                 break;
4701                         case MACMATCH_MASK:
4702                                 bits +=  9;
4703                                 break;
4704                         case MPSHITTYPE_MASK:
4705                                 bits +=  3;
4706                                 break;
4707                         case FRAGMENTATION_MASK:
4708                                 bits +=  1;
4709                                 break;
4710                         }
4711
4712                 if (bits > 36) {
4713                         dev_err(adapter->pdev_dev,
4714                                 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
4715                                 " using %#x\n", tp_vlan_pri_map, bits,
4716                                 TP_VLAN_PRI_MAP_DEFAULT);
4717                         tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
4718                 }
4719         }
4720         v = tp_vlan_pri_map;
4721         t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
4722                           &v, 1, TP_VLAN_PRI_MAP);
4723
4724         /*
4725          * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
4726          * to support any of the compressed filter fields above.  Newer
4727          * versions of the firmware do this automatically but it doesn't hurt
4728          * to set it here.  Meanwhile, we do _not_ need to set Lookup Every
4729          * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
4730          * since the firmware automatically turns this on and off when we have
4731          * a non-zero number of filters active (since it does have a
4732          * performance impact).
4733          */
4734         if (tp_vlan_pri_map)
4735                 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
4736                                  FIVETUPLELOOKUP_MASK,
4737                                  FIVETUPLELOOKUP_MASK);
4738
4739         /*
4740          * Tweak some settings.
4741          */
4742         t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
4743                      RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
4744                      PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
4745                      KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
4746
4747         /*
4748          * Get basic stuff going by issuing the Firmware Initialize command.
4749          * Note that this _must_ be after all PFVF commands ...
4750          */
4751         ret = t4_fw_initialize(adapter, adapter->mbox);
4752         if (ret < 0)
4753                 goto bye;
4754
4755         /*
4756          * Return successfully!
4757          */
4758         dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
4759                  "driver parameters\n");
4760         return 0;
4761
4762         /*
4763          * Something bad happened.  Return the error ...
4764          */
4765 bye:
4766         return ret;
4767 }
4768
4769 /*
4770  * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4771  */
4772 static int adap_init0(struct adapter *adap)
4773 {
4774         int ret;
4775         u32 v, port_vec;
4776         enum dev_state state;
4777         u32 params[7], val[7];
4778         struct fw_caps_config_cmd caps_cmd;
4779         int reset = 1, j;
4780
4781         /*
4782          * Contact FW, advertising Master capability (and potentially forcing
4783          * ourselves as the Master PF if our module parameter force_init is
4784          * set).
4785          */
4786         ret = t4_fw_hello(adap, adap->mbox, adap->fn,
4787                           force_init ? MASTER_MUST : MASTER_MAY,
4788                           &state);
4789         if (ret < 0) {
4790                 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4791                         ret);
4792                 return ret;
4793         }
4794         if (ret == adap->mbox)
4795                 adap->flags |= MASTER_PF;
4796         if (force_init && state == DEV_STATE_INIT)
4797                 state = DEV_STATE_UNINIT;
4798
4799         /*
4800          * If we're the Master PF Driver and the device is uninitialized,
4801          * then let's consider upgrading the firmware ...  (We always want
4802          * to check the firmware version number in order to A. get it for
4803          * later reporting and B. to warn if the currently loaded firmware
4804          * is excessively mismatched relative to the driver.)
4805          */
4806         ret = t4_check_fw_version(adap);
4807         if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
4808                 if (ret == -EINVAL || ret > 0) {
4809                         if (upgrade_fw(adap) >= 0) {
4810                                 /*
4811                                  * Note that the chip was reset as part of the
4812                                  * firmware upgrade so we don't reset it again
4813                                  * below and grab the new firmware version.
4814                                  */
4815                                 reset = 0;
4816                                 ret = t4_check_fw_version(adap);
4817                         }
4818                 }
4819                 if (ret < 0)
4820                         return ret;
4821         }
4822
4823         /*
4824          * Grab VPD parameters.  This should be done after we establish a
4825          * connection to the firmware since some of the VPD parameters
4826          * (notably the Core Clock frequency) are retrieved via requests to
4827          * the firmware.  On the other hand, we need these fairly early on
4828          * so we do this right after getting ahold of the firmware.
4829          */
4830         ret = get_vpd_params(adap, &adap->params.vpd);
4831         if (ret < 0)
4832                 goto bye;
4833
4834         /*
4835          * Find out what ports are available to us.  Note that we need to do
4836          * this before calling adap_init0_no_config() since it needs nports
4837          * and portvec ...
4838          */
4839         v =
4840             FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4841             FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
4842         ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
4843         if (ret < 0)
4844                 goto bye;
4845
4846         adap->params.nports = hweight32(port_vec);
4847         adap->params.portvec = port_vec;
4848
4849         /*
4850          * If the firmware is initialized already (and we're not forcing a
4851          * master initialization), note that we're living with existing
4852          * adapter parameters.  Otherwise, it's time to try initializing the
4853          * adapter ...
4854          */
4855         if (state == DEV_STATE_INIT) {
4856                 dev_info(adap->pdev_dev, "Coming up as %s: "\
4857                          "Adapter already initialized\n",
4858                          adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
4859                 adap->flags |= USING_SOFT_PARAMS;
4860         } else {
4861                 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4862                          "Initializing adapter\n");
4863
4864                 /*
4865                  * If the firmware doesn't support Configuration
4866                  * Files warn user and exit,
4867                  */
4868                 if (ret < 0)
4869                         dev_warn(adap->pdev_dev, "Firmware doesn't support "
4870                                  "configuration file.\n");
4871                 if (force_old_init)
4872                         ret = adap_init0_no_config(adap, reset);
4873                 else {
4874                         /*
4875                          * Find out whether we're dealing with a version of
4876                          * the firmware which has configuration file support.
4877                          */
4878                         params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4879                                      FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4880                         ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
4881                                               params, val);
4882
4883                         /*
4884                          * If the firmware doesn't support Configuration
4885                          * Files, use the old Driver-based, hard-wired
4886                          * initialization.  Otherwise, try using the
4887                          * Configuration File support and fall back to the
4888                          * Driver-based initialization if there's no
4889                          * Configuration File found.
4890                          */
4891                         if (ret < 0)
4892                                 ret = adap_init0_no_config(adap, reset);
4893                         else {
4894                                 /*
4895                                  * The firmware provides us with a memory
4896                                  * buffer where we can load a Configuration
4897                                  * File from the host if we want to override
4898                                  * the Configuration File in flash.
4899                                  */
4900
4901                                 ret = adap_init0_config(adap, reset);
4902                                 if (ret == -ENOENT) {
4903                                         dev_info(adap->pdev_dev,
4904                                             "No Configuration File present "
4905                                             "on adapter.  Using hard-wired "
4906                                             "configuration parameters.\n");
4907                                         ret = adap_init0_no_config(adap, reset);
4908                                 }
4909                         }
4910                 }
4911                 if (ret < 0) {
4912                         dev_err(adap->pdev_dev,
4913                                 "could not initialize adapter, error %d\n",
4914                                 -ret);
4915                         goto bye;
4916                 }
4917         }
4918
4919         /*
4920          * If we're living with non-hard-coded parameters (either from a
4921          * Firmware Configuration File or values programmed by a different PF
4922          * Driver), give the SGE code a chance to pull in anything that it
4923          * needs ...  Note that this must be called after we retrieve our VPD
4924          * parameters in order to know how to convert core ticks to seconds.
4925          */
4926         if (adap->flags & USING_SOFT_PARAMS) {
4927                 ret = t4_sge_init(adap);
4928                 if (ret < 0)
4929                         goto bye;
4930         }
4931
4932         if (is_bypass_device(adap->pdev->device))
4933                 adap->params.bypass = 1;
4934
4935         /*
4936          * Grab some of our basic fundamental operating parameters.
4937          */
4938 #define FW_PARAM_DEV(param) \
4939         (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4940         FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4941
4942 #define FW_PARAM_PFVF(param) \
4943         FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4944         FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)|  \
4945         FW_PARAMS_PARAM_Y(0) | \
4946         FW_PARAMS_PARAM_Z(0)
4947
4948         params[0] = FW_PARAM_PFVF(EQ_START);
4949         params[1] = FW_PARAM_PFVF(L2T_START);
4950         params[2] = FW_PARAM_PFVF(L2T_END);
4951         params[3] = FW_PARAM_PFVF(FILTER_START);
4952         params[4] = FW_PARAM_PFVF(FILTER_END);
4953         params[5] = FW_PARAM_PFVF(IQFLINT_START);
4954         ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
4955         if (ret < 0)
4956                 goto bye;
4957         adap->sge.egr_start = val[0];
4958         adap->l2t_start = val[1];
4959         adap->l2t_end = val[2];
4960         adap->tids.ftid_base = val[3];
4961         adap->tids.nftids = val[4] - val[3] + 1;
4962         adap->sge.ingr_start = val[5];
4963
4964         /* query params related to active filter region */
4965         params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4966         params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4967         ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
4968         /* If Active filter size is set we enable establishing
4969          * offload connection through firmware work request
4970          */
4971         if ((val[0] != val[1]) && (ret >= 0)) {
4972                 adap->flags |= FW_OFLD_CONN;
4973                 adap->tids.aftid_base = val[0];
4974                 adap->tids.aftid_end = val[1];
4975         }
4976
4977         /*
4978          * Get device capabilities so we can determine what resources we need
4979          * to manage.
4980          */
4981         memset(&caps_cmd, 0, sizeof(caps_cmd));
4982         caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4983                                      FW_CMD_REQUEST | FW_CMD_READ);
4984         caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4985         ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4986                          &caps_cmd);
4987         if (ret < 0)
4988                 goto bye;
4989
4990         if (caps_cmd.ofldcaps) {
4991                 /* query offload-related parameters */
4992                 params[0] = FW_PARAM_DEV(NTID);
4993                 params[1] = FW_PARAM_PFVF(SERVER_START);
4994                 params[2] = FW_PARAM_PFVF(SERVER_END);
4995                 params[3] = FW_PARAM_PFVF(TDDP_START);
4996                 params[4] = FW_PARAM_PFVF(TDDP_END);
4997                 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4998                 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
4999                                       params, val);
5000                 if (ret < 0)
5001                         goto bye;
5002                 adap->tids.ntids = val[0];
5003                 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
5004                 adap->tids.stid_base = val[1];
5005                 adap->tids.nstids = val[2] - val[1] + 1;
5006                 /*
5007                  * Setup server filter region. Divide the availble filter
5008                  * region into two parts. Regular filters get 1/3rd and server
5009                  * filters get 2/3rd part. This is only enabled if workarond
5010                  * path is enabled.
5011                  * 1. For regular filters.
5012                  * 2. Server filter: This are special filters which are used
5013                  * to redirect SYN packets to offload queue.
5014                  */
5015                 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
5016                         adap->tids.sftid_base = adap->tids.ftid_base +
5017                                         DIV_ROUND_UP(adap->tids.nftids, 3);
5018                         adap->tids.nsftids = adap->tids.nftids -
5019                                          DIV_ROUND_UP(adap->tids.nftids, 3);
5020                         adap->tids.nftids = adap->tids.sftid_base -
5021                                                 adap->tids.ftid_base;
5022                 }
5023                 adap->vres.ddp.start = val[3];
5024                 adap->vres.ddp.size = val[4] - val[3] + 1;
5025                 adap->params.ofldq_wr_cred = val[5];
5026
5027                 adap->params.offload = 1;
5028         }
5029         if (caps_cmd.rdmacaps) {
5030                 params[0] = FW_PARAM_PFVF(STAG_START);
5031                 params[1] = FW_PARAM_PFVF(STAG_END);
5032                 params[2] = FW_PARAM_PFVF(RQ_START);
5033                 params[3] = FW_PARAM_PFVF(RQ_END);
5034                 params[4] = FW_PARAM_PFVF(PBL_START);
5035                 params[5] = FW_PARAM_PFVF(PBL_END);
5036                 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5037                                       params, val);
5038                 if (ret < 0)
5039                         goto bye;
5040                 adap->vres.stag.start = val[0];
5041                 adap->vres.stag.size = val[1] - val[0] + 1;
5042                 adap->vres.rq.start = val[2];
5043                 adap->vres.rq.size = val[3] - val[2] + 1;
5044                 adap->vres.pbl.start = val[4];
5045                 adap->vres.pbl.size = val[5] - val[4] + 1;
5046
5047                 params[0] = FW_PARAM_PFVF(SQRQ_START);
5048                 params[1] = FW_PARAM_PFVF(SQRQ_END);
5049                 params[2] = FW_PARAM_PFVF(CQ_START);
5050                 params[3] = FW_PARAM_PFVF(CQ_END);
5051                 params[4] = FW_PARAM_PFVF(OCQ_START);
5052                 params[5] = FW_PARAM_PFVF(OCQ_END);
5053                 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
5054                 if (ret < 0)
5055                         goto bye;
5056                 adap->vres.qp.start = val[0];
5057                 adap->vres.qp.size = val[1] - val[0] + 1;
5058                 adap->vres.cq.start = val[2];
5059                 adap->vres.cq.size = val[3] - val[2] + 1;
5060                 adap->vres.ocq.start = val[4];
5061                 adap->vres.ocq.size = val[5] - val[4] + 1;
5062         }
5063         if (caps_cmd.iscsicaps) {
5064                 params[0] = FW_PARAM_PFVF(ISCSI_START);
5065                 params[1] = FW_PARAM_PFVF(ISCSI_END);
5066                 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
5067                                       params, val);
5068                 if (ret < 0)
5069                         goto bye;
5070                 adap->vres.iscsi.start = val[0];
5071                 adap->vres.iscsi.size = val[1] - val[0] + 1;
5072         }
5073 #undef FW_PARAM_PFVF
5074 #undef FW_PARAM_DEV
5075
5076         /*
5077          * These are finalized by FW initialization, load their values now.
5078          */
5079         v = t4_read_reg(adap, TP_TIMER_RESOLUTION);
5080         adap->params.tp.tre = TIMERRESOLUTION_GET(v);
5081         adap->params.tp.dack_re = DELAYEDACKRESOLUTION_GET(v);
5082         t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5083         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5084                      adap->params.b_wnd);
5085
5086         /* MODQ_REQ_MAP defaults to setting queues 0-3 to chan 0-3 */
5087         for (j = 0; j < NCHAN; j++)
5088                 adap->params.tp.tx_modq[j] = j;
5089
5090         t4_read_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
5091                          &adap->filter_mode, 1,
5092                          TP_VLAN_PRI_MAP);
5093
5094         adap->flags |= FW_OK;
5095         return 0;
5096
5097         /*
5098          * Something bad happened.  If a command timed out or failed with EIO
5099          * FW does not operate within its spec or something catastrophic
5100          * happened to HW/FW, stop issuing commands.
5101          */
5102 bye:
5103         if (ret != -ETIMEDOUT && ret != -EIO)
5104                 t4_fw_bye(adap, adap->mbox);
5105         return ret;
5106 }
5107
5108 /* EEH callbacks */
5109
5110 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5111                                          pci_channel_state_t state)
5112 {
5113         int i;
5114         struct adapter *adap = pci_get_drvdata(pdev);
5115
5116         if (!adap)
5117                 goto out;
5118
5119         rtnl_lock();
5120         adap->flags &= ~FW_OK;
5121         notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5122         for_each_port(adap, i) {
5123                 struct net_device *dev = adap->port[i];
5124
5125                 netif_device_detach(dev);
5126                 netif_carrier_off(dev);
5127         }
5128         if (adap->flags & FULL_INIT_DONE)
5129                 cxgb_down(adap);
5130         rtnl_unlock();
5131         pci_disable_device(pdev);
5132 out:    return state == pci_channel_io_perm_failure ?
5133                 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5134 }
5135
5136 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5137 {
5138         int i, ret;
5139         struct fw_caps_config_cmd c;
5140         struct adapter *adap = pci_get_drvdata(pdev);
5141
5142         if (!adap) {
5143                 pci_restore_state(pdev);
5144                 pci_save_state(pdev);
5145                 return PCI_ERS_RESULT_RECOVERED;
5146         }
5147
5148         if (pci_enable_device(pdev)) {
5149                 dev_err(&pdev->dev, "cannot reenable PCI device after reset\n");
5150                 return PCI_ERS_RESULT_DISCONNECT;
5151         }
5152
5153         pci_set_master(pdev);
5154         pci_restore_state(pdev);
5155         pci_save_state(pdev);
5156         pci_cleanup_aer_uncorrect_error_status(pdev);
5157
5158         if (t4_wait_dev_ready(adap) < 0)
5159                 return PCI_ERS_RESULT_DISCONNECT;
5160         if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL))
5161                 return PCI_ERS_RESULT_DISCONNECT;
5162         adap->flags |= FW_OK;
5163         if (adap_init1(adap, &c))
5164                 return PCI_ERS_RESULT_DISCONNECT;
5165
5166         for_each_port(adap, i) {
5167                 struct port_info *p = adap2pinfo(adap, i);
5168
5169                 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5170                                   NULL, NULL);
5171                 if (ret < 0)
5172                         return PCI_ERS_RESULT_DISCONNECT;
5173                 p->viid = ret;
5174                 p->xact_addr_filt = -1;
5175         }
5176
5177         t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5178                      adap->params.b_wnd);
5179         setup_memwin(adap);
5180         if (cxgb_up(adap))
5181                 return PCI_ERS_RESULT_DISCONNECT;
5182         return PCI_ERS_RESULT_RECOVERED;
5183 }
5184
5185 static void eeh_resume(struct pci_dev *pdev)
5186 {
5187         int i;
5188         struct adapter *adap = pci_get_drvdata(pdev);
5189
5190         if (!adap)
5191                 return;
5192
5193         rtnl_lock();
5194         for_each_port(adap, i) {
5195                 struct net_device *dev = adap->port[i];
5196
5197                 if (netif_running(dev)) {
5198                         link_start(dev);
5199                         cxgb_set_rxmode(dev);
5200                 }
5201                 netif_device_attach(dev);
5202         }
5203         rtnl_unlock();
5204 }
5205
5206 static const struct pci_error_handlers cxgb4_eeh = {
5207         .error_detected = eeh_err_detected,
5208         .slot_reset     = eeh_slot_reset,
5209         .resume         = eeh_resume,
5210 };
5211
5212 static inline bool is_10g_port(const struct link_config *lc)
5213 {
5214         return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0;
5215 }
5216
5217 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
5218                              unsigned int size, unsigned int iqe_size)
5219 {
5220         q->intr_params = QINTR_TIMER_IDX(timer_idx) |
5221                          (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
5222         q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
5223         q->iqe_len = iqe_size;
5224         q->size = size;
5225 }
5226
5227 /*
5228  * Perform default configuration of DMA queues depending on the number and type
5229  * of ports we found and the number of available CPUs.  Most settings can be
5230  * modified by the admin prior to actual use.
5231  */
5232 static void cfg_queues(struct adapter *adap)
5233 {
5234         struct sge *s = &adap->sge;
5235         int i, q10g = 0, n10g = 0, qidx = 0;
5236
5237         for_each_port(adap, i)
5238                 n10g += is_10g_port(&adap2pinfo(adap, i)->link_cfg);
5239
5240         /*
5241          * We default to 1 queue per non-10G port and up to # of cores queues
5242          * per 10G port.
5243          */
5244         if (n10g)
5245                 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5246         if (q10g > netif_get_num_default_rss_queues())
5247                 q10g = netif_get_num_default_rss_queues();
5248
5249         for_each_port(adap, i) {
5250                 struct port_info *pi = adap2pinfo(adap, i);
5251
5252                 pi->first_qset = qidx;
5253                 pi->nqsets = is_10g_port(&pi->link_cfg) ? q10g : 1;
5254                 qidx += pi->nqsets;
5255         }
5256
5257         s->ethqsets = qidx;
5258         s->max_ethqsets = qidx;   /* MSI-X may lower it later */
5259
5260         if (is_offload(adap)) {
5261                 /*
5262                  * For offload we use 1 queue/channel if all ports are up to 1G,
5263                  * otherwise we divide all available queues amongst the channels
5264                  * capped by the number of available cores.
5265                  */
5266                 if (n10g) {
5267                         i = min_t(int, ARRAY_SIZE(s->ofldrxq),
5268                                   num_online_cpus());
5269                         s->ofldqsets = roundup(i, adap->params.nports);
5270                 } else
5271                         s->ofldqsets = adap->params.nports;
5272                 /* For RDMA one Rx queue per channel suffices */
5273                 s->rdmaqs = adap->params.nports;
5274         }
5275
5276         for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5277                 struct sge_eth_rxq *r = &s->ethrxq[i];
5278
5279                 init_rspq(&r->rspq, 0, 0, 1024, 64);
5280                 r->fl.size = 72;
5281         }
5282
5283         for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5284                 s->ethtxq[i].q.size = 1024;
5285
5286         for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5287                 s->ctrlq[i].q.size = 512;
5288
5289         for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
5290                 s->ofldtxq[i].q.size = 1024;
5291
5292         for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
5293                 struct sge_ofld_rxq *r = &s->ofldrxq[i];
5294
5295                 init_rspq(&r->rspq, 0, 0, 1024, 64);
5296                 r->rspq.uld = CXGB4_ULD_ISCSI;
5297                 r->fl.size = 72;
5298         }
5299
5300         for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
5301                 struct sge_ofld_rxq *r = &s->rdmarxq[i];
5302
5303                 init_rspq(&r->rspq, 0, 0, 511, 64);
5304                 r->rspq.uld = CXGB4_ULD_RDMA;
5305                 r->fl.size = 72;
5306         }
5307
5308         init_rspq(&s->fw_evtq, 6, 0, 512, 64);
5309         init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
5310 }
5311
5312 /*
5313  * Reduce the number of Ethernet queues across all ports to at most n.
5314  * n provides at least one queue per port.
5315  */
5316 static void reduce_ethqs(struct adapter *adap, int n)
5317 {
5318         int i;
5319         struct port_info *pi;
5320
5321         while (n < adap->sge.ethqsets)
5322                 for_each_port(adap, i) {
5323                         pi = adap2pinfo(adap, i);
5324                         if (pi->nqsets > 1) {
5325                                 pi->nqsets--;
5326                                 adap->sge.ethqsets--;
5327                                 if (adap->sge.ethqsets <= n)
5328                                         break;
5329                         }
5330                 }
5331
5332         n = 0;
5333         for_each_port(adap, i) {
5334                 pi = adap2pinfo(adap, i);
5335                 pi->first_qset = n;
5336                 n += pi->nqsets;
5337         }
5338 }
5339
5340 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5341 #define EXTRA_VECS 2
5342
5343 static int enable_msix(struct adapter *adap)
5344 {
5345         int ofld_need = 0;
5346         int i, err, want, need;
5347         struct sge *s = &adap->sge;
5348         unsigned int nchan = adap->params.nports;
5349         struct msix_entry entries[MAX_INGQ + 1];
5350
5351         for (i = 0; i < ARRAY_SIZE(entries); ++i)
5352                 entries[i].entry = i;
5353
5354         want = s->max_ethqsets + EXTRA_VECS;
5355         if (is_offload(adap)) {
5356                 want += s->rdmaqs + s->ofldqsets;
5357                 /* need nchan for each possible ULD */
5358                 ofld_need = 2 * nchan;
5359         }
5360         need = adap->params.nports + EXTRA_VECS + ofld_need;
5361
5362         while ((err = pci_enable_msix(adap->pdev, entries, want)) >= need)
5363                 want = err;
5364
5365         if (!err) {
5366                 /*
5367                  * Distribute available vectors to the various queue groups.
5368                  * Every group gets its minimum requirement and NIC gets top
5369                  * priority for leftovers.
5370                  */
5371                 i = want - EXTRA_VECS - ofld_need;
5372                 if (i < s->max_ethqsets) {
5373                         s->max_ethqsets = i;
5374                         if (i < s->ethqsets)
5375                                 reduce_ethqs(adap, i);
5376                 }
5377                 if (is_offload(adap)) {
5378                         i = want - EXTRA_VECS - s->max_ethqsets;
5379                         i -= ofld_need - nchan;
5380                         s->ofldqsets = (i / nchan) * nchan;  /* round down */
5381                 }
5382                 for (i = 0; i < want; ++i)
5383                         adap->msix_info[i].vec = entries[i].vector;
5384         } else if (err > 0)
5385                 dev_info(adap->pdev_dev,
5386                          "only %d MSI-X vectors left, not using MSI-X\n", err);
5387         return err;
5388 }
5389
5390 #undef EXTRA_VECS
5391
5392 static int init_rss(struct adapter *adap)
5393 {
5394         unsigned int i, j;
5395
5396         for_each_port(adap, i) {
5397                 struct port_info *pi = adap2pinfo(adap, i);
5398
5399                 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5400                 if (!pi->rss)
5401                         return -ENOMEM;
5402                 for (j = 0; j < pi->rss_size; j++)
5403                         pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
5404         }
5405         return 0;
5406 }
5407
5408 static void print_port_info(const struct net_device *dev)
5409 {
5410         static const char *base[] = {
5411                 "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4",
5412                 "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4"
5413         };
5414
5415         char buf[80];
5416         char *bufp = buf;
5417         const char *spd = "";
5418         const struct port_info *pi = netdev_priv(dev);
5419         const struct adapter *adap = pi->adapter;
5420
5421         if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
5422                 spd = " 2.5 GT/s";
5423         else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
5424                 spd = " 5 GT/s";
5425
5426         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5427                 bufp += sprintf(bufp, "100/");
5428         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5429                 bufp += sprintf(bufp, "1000/");
5430         if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
5431                 bufp += sprintf(bufp, "10G/");
5432         if (bufp != buf)
5433                 --bufp;
5434         sprintf(bufp, "BASE-%s", base[pi->port_type]);
5435
5436         netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5437                     adap->params.vpd.id,
5438                     CHELSIO_CHIP_RELEASE(adap->params.rev), buf,
5439                     is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
5440                     (adap->flags & USING_MSIX) ? " MSI-X" :
5441                     (adap->flags & USING_MSI) ? " MSI" : "");
5442         netdev_info(dev, "S/N: %s, E/C: %s\n",
5443                     adap->params.vpd.sn, adap->params.vpd.ec);
5444 }
5445
5446 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
5447 {
5448         pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
5449 }
5450
5451 /*
5452  * Free the following resources:
5453  * - memory used for tables
5454  * - MSI/MSI-X
5455  * - net devices
5456  * - resources FW is holding for us
5457  */
5458 static void free_some_resources(struct adapter *adapter)
5459 {
5460         unsigned int i;
5461
5462         t4_free_mem(adapter->l2t);
5463         t4_free_mem(adapter->tids.tid_tab);
5464         disable_msi(adapter);
5465
5466         for_each_port(adapter, i)
5467                 if (adapter->port[i]) {
5468                         kfree(adap2pinfo(adapter, i)->rss);
5469                         free_netdev(adapter->port[i]);
5470                 }
5471         if (adapter->flags & FW_OK)
5472                 t4_fw_bye(adapter, adapter->fn);
5473 }
5474
5475 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5476 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5477                    NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5478 #define SEGMENT_SIZE 128
5479
5480 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5481 {
5482         int func, i, err, s_qpp, qpp, num_seg;
5483         struct port_info *pi;
5484         bool highdma = false;
5485         struct adapter *adapter = NULL;
5486 #ifdef CONFIG_PCI_IOV
5487         int max_no_pf;
5488 #endif
5489
5490         printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5491
5492         err = pci_request_regions(pdev, KBUILD_MODNAME);
5493         if (err) {
5494                 /* Just info, some other driver may have claimed the device. */
5495                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5496                 return err;
5497         }
5498
5499         /* We control everything through one PF */
5500         func = PCI_FUNC(pdev->devfn);
5501         if (func != ent->driver_data) {
5502                 pci_save_state(pdev);        /* to restore SR-IOV later */
5503                 goto sriov;
5504         }
5505
5506         err = pci_enable_device(pdev);
5507         if (err) {
5508                 dev_err(&pdev->dev, "cannot enable PCI device\n");
5509                 goto out_release_regions;
5510         }
5511
5512         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5513                 highdma = true;
5514                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5515                 if (err) {
5516                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5517                                 "coherent allocations\n");
5518                         goto out_disable_device;
5519                 }
5520         } else {
5521                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5522                 if (err) {
5523                         dev_err(&pdev->dev, "no usable DMA configuration\n");
5524                         goto out_disable_device;
5525                 }
5526         }
5527
5528         pci_enable_pcie_error_reporting(pdev);
5529         enable_pcie_relaxed_ordering(pdev);
5530         pci_set_master(pdev);
5531         pci_save_state(pdev);
5532
5533         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5534         if (!adapter) {
5535                 err = -ENOMEM;
5536                 goto out_disable_device;
5537         }
5538
5539         adapter->regs = pci_ioremap_bar(pdev, 0);
5540         if (!adapter->regs) {
5541                 dev_err(&pdev->dev, "cannot map device registers\n");
5542                 err = -ENOMEM;
5543                 goto out_free_adapter;
5544         }
5545
5546         adapter->pdev = pdev;
5547         adapter->pdev_dev = &pdev->dev;
5548         adapter->mbox = func;
5549         adapter->fn = func;
5550         adapter->msg_enable = dflt_msg_enable;
5551         memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5552
5553         spin_lock_init(&adapter->stats_lock);
5554         spin_lock_init(&adapter->tid_release_lock);
5555
5556         INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5557         INIT_WORK(&adapter->db_full_task, process_db_full);
5558         INIT_WORK(&adapter->db_drop_task, process_db_drop);
5559
5560         err = t4_prep_adapter(adapter);
5561         if (err)
5562                 goto out_unmap_bar0;
5563
5564         if (!is_t4(adapter->chip)) {
5565                 s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
5566                 qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
5567                       SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
5568                 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5569
5570                 /* Each segment size is 128B. Write coalescing is enabled only
5571                  * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5572                  * queue is less no of segments that can be accommodated in
5573                  * a page size.
5574                  */
5575                 if (qpp > num_seg) {
5576                         dev_err(&pdev->dev,
5577                                 "Incorrect number of egress queues per page\n");
5578                         err = -EINVAL;
5579                         goto out_unmap_bar0;
5580                 }
5581                 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5582                 pci_resource_len(pdev, 2));
5583                 if (!adapter->bar2) {
5584                         dev_err(&pdev->dev, "cannot map device bar2 region\n");
5585                         err = -ENOMEM;
5586                         goto out_unmap_bar0;
5587                 }
5588         }
5589
5590         setup_memwin(adapter);
5591         err = adap_init0(adapter);
5592         setup_memwin_rdma(adapter);
5593         if (err)
5594                 goto out_unmap_bar;
5595
5596         for_each_port(adapter, i) {
5597                 struct net_device *netdev;
5598
5599                 netdev = alloc_etherdev_mq(sizeof(struct port_info),
5600                                            MAX_ETH_QSETS);
5601                 if (!netdev) {
5602                         err = -ENOMEM;
5603                         goto out_free_dev;
5604                 }
5605
5606                 SET_NETDEV_DEV(netdev, &pdev->dev);
5607
5608                 adapter->port[i] = netdev;
5609                 pi = netdev_priv(netdev);
5610                 pi->adapter = adapter;
5611                 pi->xact_addr_filt = -1;
5612                 pi->port_id = i;
5613                 netdev->irq = pdev->irq;
5614
5615                 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5616                         NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5617                         NETIF_F_RXCSUM | NETIF_F_RXHASH |
5618                         NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
5619                 if (highdma)
5620                         netdev->hw_features |= NETIF_F_HIGHDMA;
5621                 netdev->features |= netdev->hw_features;
5622                 netdev->vlan_features = netdev->features & VLAN_FEAT;
5623
5624                 netdev->priv_flags |= IFF_UNICAST_FLT;
5625
5626                 netdev->netdev_ops = &cxgb4_netdev_ops;
5627                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
5628         }
5629
5630         pci_set_drvdata(pdev, adapter);
5631
5632         if (adapter->flags & FW_OK) {
5633                 err = t4_port_init(adapter, func, func, 0);
5634                 if (err)
5635                         goto out_free_dev;
5636         }
5637
5638         /*
5639          * Configure queues and allocate tables now, they can be needed as
5640          * soon as the first register_netdev completes.
5641          */
5642         cfg_queues(adapter);
5643
5644         adapter->l2t = t4_init_l2t();
5645         if (!adapter->l2t) {
5646                 /* We tolerate a lack of L2T, giving up some functionality */
5647                 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5648                 adapter->params.offload = 0;
5649         }
5650
5651         if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
5652                 dev_warn(&pdev->dev, "could not allocate TID table, "
5653                          "continuing\n");
5654                 adapter->params.offload = 0;
5655         }
5656
5657         /* See what interrupts we'll be using */
5658         if (msi > 1 && enable_msix(adapter) == 0)
5659                 adapter->flags |= USING_MSIX;
5660         else if (msi > 0 && pci_enable_msi(pdev) == 0)
5661                 adapter->flags |= USING_MSI;
5662
5663         err = init_rss(adapter);
5664         if (err)
5665                 goto out_free_dev;
5666
5667         /*
5668          * The card is now ready to go.  If any errors occur during device
5669          * registration we do not fail the whole card but rather proceed only
5670          * with the ports we manage to register successfully.  However we must
5671          * register at least one net device.
5672          */
5673         for_each_port(adapter, i) {
5674                 pi = adap2pinfo(adapter, i);
5675                 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5676                 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5677
5678                 err = register_netdev(adapter->port[i]);
5679                 if (err)
5680                         break;
5681                 adapter->chan_map[pi->tx_chan] = i;
5682                 print_port_info(adapter->port[i]);
5683         }
5684         if (i == 0) {
5685                 dev_err(&pdev->dev, "could not register any net devices\n");
5686                 goto out_free_dev;
5687         }
5688         if (err) {
5689                 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5690                 err = 0;
5691         }
5692
5693         if (cxgb4_debugfs_root) {
5694                 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5695                                                            cxgb4_debugfs_root);
5696                 setup_debugfs(adapter);
5697         }
5698
5699         /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5700         pdev->needs_freset = 1;
5701
5702         if (is_offload(adapter))
5703                 attach_ulds(adapter);
5704
5705 sriov:
5706 #ifdef CONFIG_PCI_IOV
5707         max_no_pf = is_t4(adapter->chip) ? NUM_OF_PF_WITH_SRIOV_T4 :
5708                         NUM_OF_PF_WITH_SRIOV_T5;
5709
5710         if (func < max_no_pf && num_vf[func] > 0)
5711                 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
5712                         dev_info(&pdev->dev,
5713                                  "instantiated %u virtual functions\n",
5714                                  num_vf[func]);
5715 #endif
5716         return 0;
5717
5718  out_free_dev:
5719         free_some_resources(adapter);
5720  out_unmap_bar:
5721         if (!is_t4(adapter->chip))
5722                 iounmap(adapter->bar2);
5723  out_unmap_bar0:
5724         iounmap(adapter->regs);
5725  out_free_adapter:
5726         kfree(adapter);
5727  out_disable_device:
5728         pci_disable_pcie_error_reporting(pdev);
5729         pci_disable_device(pdev);
5730  out_release_regions:
5731         pci_release_regions(pdev);
5732         pci_set_drvdata(pdev, NULL);
5733         return err;
5734 }
5735
5736 static void remove_one(struct pci_dev *pdev)
5737 {
5738         struct adapter *adapter = pci_get_drvdata(pdev);
5739
5740 #ifdef CONFIG_PCI_IOV
5741         pci_disable_sriov(pdev);
5742
5743 #endif
5744
5745         if (adapter) {
5746                 int i;
5747
5748                 if (is_offload(adapter))
5749                         detach_ulds(adapter);
5750
5751                 for_each_port(adapter, i)
5752                         if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5753                                 unregister_netdev(adapter->port[i]);
5754
5755                 if (adapter->debugfs_root)
5756                         debugfs_remove_recursive(adapter->debugfs_root);
5757
5758                 /* If we allocated filters, free up state associated with any
5759                  * valid filters ...
5760                  */
5761                 if (adapter->tids.ftid_tab) {
5762                         struct filter_entry *f = &adapter->tids.ftid_tab[0];
5763                         for (i = 0; i < (adapter->tids.nftids +
5764                                         adapter->tids.nsftids); i++, f++)
5765                                 if (f->valid)
5766                                         clear_filter(adapter, f);
5767                 }
5768
5769                 if (adapter->flags & FULL_INIT_DONE)
5770                         cxgb_down(adapter);
5771
5772                 free_some_resources(adapter);
5773                 iounmap(adapter->regs);
5774                 if (!is_t4(adapter->chip))
5775                         iounmap(adapter->bar2);
5776                 kfree(adapter);
5777                 pci_disable_pcie_error_reporting(pdev);
5778                 pci_disable_device(pdev);
5779                 pci_release_regions(pdev);
5780                 pci_set_drvdata(pdev, NULL);
5781         } else
5782                 pci_release_regions(pdev);
5783 }
5784
5785 static struct pci_driver cxgb4_driver = {
5786         .name     = KBUILD_MODNAME,
5787         .id_table = cxgb4_pci_tbl,
5788         .probe    = init_one,
5789         .remove   = remove_one,
5790         .err_handler = &cxgb4_eeh,
5791 };
5792
5793 static int __init cxgb4_init_module(void)
5794 {
5795         int ret;
5796
5797         workq = create_singlethread_workqueue("cxgb4");
5798         if (!workq)
5799                 return -ENOMEM;
5800
5801         /* Debugfs support is optional, just warn if this fails */
5802         cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5803         if (!cxgb4_debugfs_root)
5804                 pr_warn("could not create debugfs entry, continuing\n");
5805
5806         ret = pci_register_driver(&cxgb4_driver);
5807         if (ret < 0)
5808                 debugfs_remove(cxgb4_debugfs_root);
5809         return ret;
5810 }
5811
5812 static void __exit cxgb4_cleanup_module(void)
5813 {
5814         pci_unregister_driver(&cxgb4_driver);
5815         debugfs_remove(cxgb4_debugfs_root);  /* NULL ok */
5816         flush_workqueue(workq);
5817         destroy_workqueue(workq);
5818 }
5819
5820 module_init(cxgb4_init_module);
5821 module_exit(cxgb4_cleanup_module);